Can AI Hack You? What’s Real, What’s Hype and How to Protect Yourself

Yes—criminals can use AI to help compromise accounts, attack software and deceive people. But AI does not give someone automatic access to your phone, bank account or computer. An attacker still needs a way in: stolen login details, a security flaw, malicious software, excessive permissions or a person persuaded to approve the wrong thing.

AI can make parts of that process faster and more convincing. It can also introduce new risks when assistants receive access to private files and connected services. Understanding those entry points is more useful than imagining an all-powerful robot hacker.

Evidence reviewed September 20, 2026. This article distinguishes documented misuse, capability assessments and illustrative scenarios.

Short answer: AI can assist hacking and fraud, and some systems can automate sequences of technical actions. That does not mean they can reliably break into any target. For individuals, the practical priorities remain protecting accounts, verifying unexpected requests, updating devices and limiting what connected apps can access.

Table of Contents

• What Does “AI Hacking” Actually Mean?
• How Criminals Can Use AI Against You
• Can AI Hack Someone Without Human Help?
• Can AI Crack Your Password?
• Can AI Hack Your Phone or Bank Account?
• Can Your Own AI Assistant Become a Security Risk?
• What Is Real—and What Is Hype?
• How to Protect Yourself
• What to Do If You Think You Have Been Hacked
• What Small Businesses Should Do
• Frequently Asked Questions
• Sources and Methodology

What Does “AI Hacking” Actually Mean?

The phrase combines several different activities:

• AI-assisted deception: Generating convincing messages, fake identities, images or voices.
• AI-assisted technical attacks: Helping an attacker research systems, analyze software or develop malicious code.
• Agentic attacks: Connecting a model to tools so it can carry out multiple actions toward an objective.
• Attacks on AI systems: Manipulating an assistant or exploiting the software and permissions surrounding it.

These categories overlap, but they are not interchangeable. A scammer using a cloned voice to request money has not necessarily broken into a device. A model finding a software flaw has not necessarily compromised a live service.

In its assessment of AI threats through 2027, the UK's National Cyber Security Centre expects AI to make elements of cyber intrusions more efficient and effective. It highlights assistance with activities including reconnaissance, vulnerability research, social engineering and malware generation.

That is a serious change in attackers' capabilities. It is not a claim that every attack succeeds or that conventional security protections have stopped working.

How Criminals Can Use AI Against You

1. More Convincing Messages and Impersonation

A suspicious message no longer needs to contain spelling mistakes or awkward grammar. AI can help produce polished text and tailor its tone to a particular audience.

The FBI has warned about criminals using generative AI to make fraud more believable and operate at greater scale, including through synthetic text, images, audio and video.

The practical consequence is simple: professional wording, a familiar face or a recognizable voice should not be treated as sufficient proof of identity.

2. Personalized Social Engineering

Consider an illustrative scenario: a freelancer receives an apparent client message referencing a real project and asking them to open a “revised invoice.” The details make the request feel familiar, but the attachment or destination is malicious.

AI may help prepare or personalize such a message. The point of failure is still the action it persuades the recipient to take.

A message can also come from a genuinely compromised account. Recognizing the sender's address is helpful, but it does not make an unusual request safe.

3. Faster Technical Work

AI can assist with analyzing code, explaining errors and identifying possible weaknesses. Those capabilities can benefit defenders or help attackers.

The NCSC's 2026 discussion of frontier AI emphasizes that the relevant capability often comes from a complete system: a model combined with tools, workflows and access.

A model producing a plausible technical suggestion is different from a system successfully carrying out an intrusion. Testing, access and the target's defenses still matter.

4. Faster Attacks Against Unpatched Systems

Some attacks exploit flaws for which a security update already exists. AI assistance could make researching and using those flaws faster.

The NCSC assessment warns that AI will increase pressure on the interval between a vulnerability becoming known and an attacker exploiting it.

For a reader or small business, the implication is practical: repeatedly postponing security updates can leave an avoidable opening.

Can AI Hack Someone Without Human Help?

AI systems can automate parts of an attack and, in some settings, sequences of actions. It would be misleading to say they always need a human to direct every step.

But “autonomous” does not mean unlimited. Someone may still have selected the objective, supplied tools, configured the environment or provided initial access. A demonstration can also involve deliberately vulnerable systems or other favorable conditions.

When you see a headline about an AI hacker, ask:

• Was this a real incident, a controlled evaluation or a vendor demonstration?
• Did the system start with credentials or access already supplied?
• Was the weakness known, deliberately planted or newly discovered?
• Did a person intervene when the system became stuck?
• Did the result involve a complete compromise or one successful step?

The NCSC's August 2026 guidance on agentic AI treats systems that can take actions as a security concern requiring safeguards and oversight. The right response is to examine actual capabilities and permissions, rather than assume either that autonomous attacks are impossible or that AI can defeat every defense.

Can AI Crack Your Password?

AI does not make every password instantly recoverable. Claims about “cracking a password in seconds” are incomplete unless they explain the password, the attack conditions and how the service stores or protects credentials.

There is also an important difference between guessing a password through a website's login screen and attempting to recover passwords from stolen password data. A headline about one situation may say little about the other.

For personal protection, you do not need to determine whether the attacker uses AI. A reused password can expose multiple accounts after one breach, while a fake sign-in page may persuade someone to hand over even a strong password.

Use passkeys where available. The NCSC recommends passkeys as a phishing-resistant way to sign in. Where you use passwords, make them unique and store them in a reputable password manager. Add multifactor authentication where available.

Passkeys improve protection against credential phishing, but they do not make a compromised device, insecure account-recovery process or fraudulent payment request harmless.

Can AI Hack Your Phone or Bank Account?

It can help an attacker pursue those targets, but knowing your name or phone number does not automatically unlock them.

Possible routes include deceiving you into sharing credentials, persuading you to install malicious software, exploiting a device vulnerability or abusing an account-recovery process. These are different mechanisms with different defenses.

Financial harm can also happen without an account takeover. If an impersonator persuades you to authorize a transfer, the transaction may use your legitimate access.

That is why “my account has a strong password” is not a complete defense against scams. You also need to verify who is requesting the action and why.

Can Your Own AI Assistant Become a Security Risk?

Yes, especially when it can read private information and take actions in connected services. An assistant that only drafts text has a different risk profile from one that can access email, change files or send information elsewhere.

One concern is prompt injection: an attacker places instructions in content the assistant encounters, such as a web page, email or document, and tries to make it treat those instructions as authoritative.

For example, a malicious document might attempt to redirect an assistant away from summarizing the document and toward disclosing unrelated information. Whether that attempt succeeds depends on the system's design, permissions and safeguards.

The NCSC warns that prompt injection is a distinct security challenge. It should not be treated as something a simple wording rule can reliably eliminate.

Practical rule: Give an assistant only the access needed for its task. Prefer read-only access where sufficient, and retain approval steps for sending sensitive information, changing account settings or making consequential transactions.

You should also review connected services periodically. An integration you no longer use does not need continuing access to your data.

What Is Real—and What Is Hype?

Claim	A More Accurate Reading
AI can make scams more convincing.	Supported by public warnings about synthetic text, audio and video.
AI can help find software weaknesses.	Yes, but discovering a possible flaw is not the same as successfully compromising a target.
AI can hack any phone instantly.	An unsupported blanket claim. Access, vulnerabilities and defenses still matter.
A convincing voice proves who is calling.	No. Verify consequential requests through a separate, trusted channel.
Every sophisticated scam uses AI.	No. The quality of a scam does not establish which tools produced it.
An AI assistant can never act outside its task.	Do not assume this. Limit permissions and keep controls around consequential actions.
You need an expensive “AI-proof” security product.	No product makes that guarantee. Evaluate concrete protections rather than marketing labels.

How to Protect Yourself

1. Start With Your Main Email Account

Email often helps control access to other services through password resets. Protect it with a passkey or strong authentication, check recovery details and investigate unfamiliar sign-ins.

2. Use Unique Passwords and Strong Authentication

Use a different password for every account that still requires one. A password manager makes this manageable. Enable multifactor authentication, and use a phishing-resistant option when supported.

Do not approve an unexpected authentication request. Do not give someone a one-time security code because they claim to be support staff.

These measures align with CISA's core security guidance on passwords, authentication, phishing and updates.

3. Verify the Request, Not Just the Voice or Writing

If someone unexpectedly asks for money, credentials or sensitive files, pause. Contact the person or organization through a number or channel you already trust.

For a bank alert, open the bank's app yourself or use the number on your card. For an urgent family request, call the person back using your saved contact. Do not rely on the contact details supplied in the suspicious message.

4. Install Security Updates

Keep your operating system, browser, apps and home router supported and updated. Enable automatic updates where practical. Replace products that no longer receive security fixes.

5. Be Selective About Downloads and Permissions

Use official distribution channels for software. Be cautious about unexpected installers, browser extensions and tools promising free access to paid AI services.

Before connecting an app to email or cloud storage, check what access it requests. Permission to read selected files is different from permission to read and modify an entire account.

6. Keep Recoverable Backups

Maintain backups of important files and check that you can restore them. Include protection against deletion or encryption spreading to the backup, rather than relying solely on a continuously synchronized folder.

The NCSC's backup guidance explains why recovery arrangements matter when attackers damage or encrypt data. Backups help restore availability; they do not reverse the theft of information.

7. Protect Connected AI Tools Like Other Powerful Apps

Use the least access necessary, review activity and preserve human approval for sensitive actions. These are practical applications of the NCSC's agentic-AI security guidance.

A tool's convenience should not be the only consideration when deciding how much of your digital life it can access.

What to Do If You Think You Have Been Hacked

You do not need to prove AI was involved before responding.

1. If money was sent, contact the payment provider immediately. Explain that you suspect fraud and ask what recovery or blocking options are available.
2. Use a trusted device to secure affected accounts. Follow the service's official recovery process if you cannot sign in.
3. Change compromised or reused passwords. Start with the affected email account when it controls access to other services.
4. Review ongoing access. Sign out unfamiliar sessions and check recovery details, connected apps and email-forwarding rules.
5. Warn affected contacts. Let them know if messages from your account may be fraudulent.
6. Preserve evidence and report the incident. Keep messages and transaction details. In the United States, use the FBI's IC3 and the FTC's reporting services; elsewhere, contact the appropriate local authority.

The FTC provides guides for recovering a hacked account and responding after a scam.

If you installed suspicious software or granted remote access, stop using that device for sensitive activity until it has been assessed. For a work account or device, notify your security or IT team promptly.

What Small Businesses Should Do

Businesses need procedures that remain effective even when an impersonation sounds convincing.

• Independently verify changes to supplier bank details.
• Require a second approval for consequential payments.
• Protect email, administrator and cloud accounts with strong authentication.
• Keep software updated and maintain tested backups.
• Limit staff and AI integrations to the access their work requires.
• Give employees a clear way to report suspicious requests without blame.

A useful payment rule is: an email or call alone cannot authorize a change to bank details. The verification process should use established contact information, not information included in the change request.

Frequently Asked Questions

Can AI hack me just because it knows my name?

No. Knowing your name does not automatically provide access to your accounts. Personal information can make impersonation more convincing, which is why you should verify unexpected requests independently.

Can someone hack me using an AI chatbot?

An attacker may use AI to assist parts of an attack. A chatbot's availability does not guarantee success: the attacker still needs an effective route into the target or a way to deceive the victim.

Can an AI voice clone steal my money?

A cloned voice can support impersonation and persuade someone to authorize a payment. The voice itself does not automatically control a bank account. Verify financial requests through a separate trusted channel.

Can AI bypass two-factor authentication?

AI is not a universal bypass. Criminals may instead try to trick users into sharing codes or approving requests, abuse account recovery or compromise a device. Phishing-resistant authentication reduces important risks, but no single measure protects every part of an account.

Does a VPN protect me from AI hacking?

A VPN does not prevent you from entering credentials on a fake website, installing malicious software or authorizing a fraudulent payment. It should not be treated as a complete defense against account compromise or scams.

Can I tell whether a scam message was written by AI?

Usually you cannot establish its origin from the wording alone. Focus on the requested action, the destination and independent verification rather than trying to detect AI writing.

Can AI protect me from hackers too?

AI can also help defenders analyze software and security information. Its defensive value depends on the system and how it is used. It complements account protection, patching and recovery planning rather than making them unnecessary.

The Verdict

AI can increase an attacker's speed, reach and persuasiveness. It does not make every account defenseless.

The most useful response is to protect the routes attackers exploit: sign-ins, software flaws, misleading requests and excessive access. Secure your main email, use strong authentication, verify consequential requests independently and keep devices updated.

You do not have to identify which model a criminal used. You need controls that still work when a message looks professional, a voice sounds familiar or a connected assistant makes a mistake.

Sources and Methodology

This article draws on public guidance and assessments from the FBI, FTC, CISA and UK National Cyber Security Centre. It distinguishes criminal misuse, technical capability and forecasts. Illustrative scenarios are not presented as documented incidents.

It does not estimate what percentage of cybercrime uses AI. Ordinary cybercrime totals cannot be treated as AI-specific losses without supporting attribution.

• FBI: Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud
• NCSC: Impact of AI on Cyber Threat From Now to 2027
• NCSC: Why Cyber Defenders Need to Be Ready for Frontier AI
• NCSC: Managing the Cyber Risk of Agentic AI
• NCSC: Prompt Injection Is Not SQL Injection
• NCSC: Passkeys—What You Need to Know
• CISA: Small and Medium-Sized Business Security Resources
• NCSC: Ransomware-Resistant Backups
• FTC: Recovering a Hacked Email or Social Media Account
• FTC: What to Do If You Were Scammed

Continue Reading

• Claude Mythos and Zero-Day Vulnerability Discovery
• Claude Mythos: AI and Cybersecurity
• What Is a Hallucination in AI?
• Is ChatGPT Getting Worse? What the Data and Users Show
• AI Ethics: Core Principles and Real Risks

Will AI Replace Radiologists? What Changes—and What Still Needs a Doctor

AI can automate important parts of radiology, but that does not mean it can replace the entire radiologist. Finding an abnormality on an image is one task. Deciding what it means for a particular patient, resolving uncertainty, recommending the next step and performing image-guided procedures involve a much broader set of responsibilities.

The realistic concern is not simply whether radiologists will disappear. It is how much of their work becomes automated, how employers reorganize that work and whether productivity gains change staffing, pay and training.

Evidence reviewed September 20, 2026. Predictions below are scenarios, not guaranteed employment outcomes.

Short answer: Current evidence supports substantial automation of radiology tasks, rather than the disappearance of radiologists as a profession. However, “AI will never replace radiologists” is too confident. Some narrowly defined workflows could require less human reading, while complex interpretation, procedures, consultation and oversight remain important human responsibilities.

Table of Contents

• Will AI Replace Radiologists?
• What AI Can Already Do in Radiology
• What Clinical Research Actually Shows
• Why Radiologists Still Matter
• Diagnostic vs. Interventional Radiology
• Could AI Reduce Jobs or Change Pay?
• Will Radiologists Be Replaced by 2030?
• Should Medical Students Still Choose Radiology?
• Which Skills Will Matter Most?
• Frequently Asked Questions
• Sources and Methodology

Will AI Replace Radiologists?

AI is more likely to change the radiologist's job substantially than eliminate the whole profession in the near term. That is an assessment of the evidence available today, not a promise about every hospital or every stage of a medical career.

There are three different developments that are often described as “replacement”:

• Task automation: Software measures a lesion, identifies a suspicious finding or prepares part of a report.
• Workflow automation: A validated system handles a defined category of examinations with fewer human reading steps.
• Occupational replacement: A healthcare service operates without needing radiologists for the full range of their responsibilities.

Success at the first level does not establish the third. But task automation still matters economically: a department that needs fewer minutes of physician time per examination may eventually organize staffing differently.

Radiology is therefore neither untouched by AI nor obviously doomed. It is a specialty where automation can be powerful, but where the consequences depend on how technology is validated and deployed.

For the wider medical-career comparison, see Which Medical Specialties Are Safest From AI?

What AI Can Already Do in Radiology

Radiology uses digital images and repeatable workflows, creating many opportunities for software assistance. However, an AI product's capabilities depend on its specific design, validation and intended use.

The FDA's AI-enabled medical-device list includes numerous radiology products. Their regulatory status applies to defined uses; it is not authorization for a general-purpose replacement for a physician.

Task	What AI Can Help Do	What Still Needs Attention
Detection	Flag findings such as suspicious lesions or other targeted abnormalities.	A tool may miss disease outside its intended target or produce false alarms.
Triage	Prioritize examinations suspected of containing urgent findings.	An unflagged examination is not necessarily normal or safe to delay.
Measurement	Measure structures, segment organs and track selected findings.	Measurements must be checked when image quality or anatomy creates difficulties.
Image processing	Support reconstruction and image-quality improvement.	Clinical usefulness depends on the examination and the validated system.
Reporting	Draft or structure report content and reduce repetitive documentation.	A fluent report can still contain an omission, incorrect statement or inconsistency.

A 2025 review in Radiology discusses these technologies alongside implementation challenges. The important distinction is between a tool performing its assigned task and an entire clinical service delivering reliable patient care.

For a broader overview of benefits and limitations, read AI in Radiology: Pros and Cons.

What Clinical Research Actually Shows

A Large Mammography Trial Shows Meaningful Automation

The Swedish Mammography Screening with Artificial Intelligence trial, known as MASAI, provides a concrete example. Its 2025 screening-performance publication reported a 29% increase in cancer detection with AI-supported screening compared with standard double reading, alongside a 44.2% reduction in screen-reading workload.

This was a particular breast-screening workflow involving radiologists. The workload measure concerned screen readings; it was not a finding that 44.2% of radiologist jobs could be removed.

What this proves: A carefully tested AI workflow can improve a clinical detection measure while reducing a defined reading workload.

What it does not prove: That every imaging service will obtain the same benefit, that radiologists are unnecessary or that the reported detection improvement establishes a mortality benefit.

Human Plus AI Is Not Automatically Better

Adding an AI suggestion does not guarantee that a clinician will use it correctly. A wrong suggestion can distract a reader, while a useful suggestion can be ignored.

In a 2025 editorial discussed by RSNA, Eric Topol and Pranav Rajpurkar argued for clearer allocation of work between AI and radiologists. Their proposals included sequential workflows and allocating different categories of cases to AI, physicians or both.

These were proposed approaches requiring clinical validation, not proof that unrestricted autonomous radiology was ready for routine use. The lesson is that workflow design and testing matter alongside the model's accuracy.

Why Radiologists Still Matter

1. A Finding Needs a Clinical Interpretation

An image abnormality is not always a diagnosis. A radiologist may need to compare previous examinations, consider treatment history, assess whether a finding is new and decide how strongly the images support different explanations.

It would be inaccurate to claim that AI can only examine pixels. Systems can also be designed to use clinical records and other information. The harder question is whether the complete system reliably handles the relevant context, uncertainty and exceptions in the setting where it is used.

2. Procedures Add Physical and Clinical Responsibilities

Interventional radiologists perform minimally invasive, image-guided procedures. Their work can include biopsies, drainage procedures and treatments involving catheters or other instruments. RadiologyInfo, the ACR and RSNA patient resource, explains this procedural role.

Performing such work involves more than identifying a target on a scan. It includes planning access, working with the patient and clinical team, and responding when the situation changes.

Robotics and AI may assist with physical procedures, so “machines cannot perform physical actions” is not a sound long-term argument. The more useful distinction is that automating a complete procedure presents additional challenges beyond interpreting an image.

3. Communication Can Change What Happens Next

A radiologist's contribution may include explaining an uncertain finding, discussing options with the referring clinician or communicating an urgent result. In patient-facing work, the conversation can also involve explaining a procedure and answering questions.

These activities are part of delivering care. A correct-looking report that arrives too late, is misunderstood or does not lead to appropriate follow-up has not completed the clinical job.

4. Safe Deployment Requires Accountability

Healthcare organizations need to know who reviews uncertain cases, responds to errors and monitors a system after deployment. A product's authorization, hospital procedures, professional obligations and payment arrangements must be considered for the particular use.

There is no need to claim that every law or insurance program permanently requires a radiologist to sign every possible AI output. The FDA evaluates devices for their intended uses; that framework does not settle all questions about liability, reimbursement or future staffing.

Diagnostic vs. Interventional Radiology: Is One Safer?

Interventional radiology has additional barriers to complete automation because it combines interpretation with physical procedures. That is a task-based assessment, not a validated ranking of future job security.

Area	Likely Automation Opportunities	Responsibilities Beyond Image Recognition
Diagnostic radiology	Detection, measurement, prioritization and report drafting.	Complex interpretation, unexpected findings, consultation and quality oversight.
Breast imaging	Screening support and selected reading-workflow changes.	Diagnostic work-up, image-guided procedures and patient communication.
Interventional radiology	Planning, navigation support, measurements and documentation.	Procedures, patient management and response to complications.

Exposure also varies within a specialty. Reading a standardized examination at high volume is a different workflow from resolving an unusual case with incomplete information.

Choosing a specialty purely because it appears “AI-proof” is risky. Training takes years, tools change and a career also depends on aptitude, working conditions and whether you enjoy the daily work.

Could AI Reduce Radiology Jobs or Change Pay?

Yes, that is possible—even if radiologists remain essential. A profession does not need to disappear for its labor market to change.

Three scenarios illustrate the uncertainty:

• Backlog reduction: AI helps the existing team handle unmet demand and provide faster results.
• Higher output expectations: Staffing stays similar, but each radiologist is expected to cover more examinations.
• Reduced hiring for some work: Where demand does not grow as quickly as productivity, an employer may need fewer additional readers than it otherwise would.

These are economic possibilities, not measured forecasts for all radiology practices. A 2025 Radiology review describes imaging volume and complexity outpacing the available workforce. That creates opportunities for assistance, but a current shortage does not guarantee unchanged employment conditions indefinitely.

The effect on pay is similarly uncertain. Faster reporting could increase the value of a physician's time, intensify competition for standardized work or change how services are reimbursed. These outcomes depend on the local market and business model.

Will Radiologists Who Use AI Replace Those Who Do Not?

The familiar slogan captures the value of adapting, but it is not a scientific employment forecast. Using AI does not automatically make someone a better radiologist.

The valuable skill is knowing when a system helps, when it fails and how to act on its output. A clinician who accepts incorrect suggestions uncritically may perform worse than one who uses the tool selectively.

Will AI Replace Radiologists by 2030?

The evidence reviewed here does not establish a credible date for the disappearance of radiologists, including 2030. It supports continued changes to tasks and workflows.

Instead of a countdown, watch for these developments:

• Prospective studies showing reliable performance across different hospitals and patient populations.
• Authorization and implementation of clearly defined autonomous workflows.
• Evidence that time savings persist after integration, error review and follow-up work are included.
• Changes in hiring, training positions and staffing—not just demonstrations or vendor claims.
• Payment and accountability arrangements that make new workflows practical.

Automating selected normal examinations would be an important change. It would still be different from replacing a department's full clinical responsibilities.

Should Medical Students Still Choose Radiology?

AI alone is not a sufficient reason to rule out radiology. It is, however, a reason to investigate what the specialty is becoming.

Before deciding, speak with practicing radiologists and trainees, observe diagnostic and procedural work, and examine training and employment conditions in the country where you plan to work.

Useful questions include:

• Do I enjoy anatomy, diagnostic uncertainty and image-based reasoning?
• Would I enjoy the work beyond producing reports?
• Does the training program teach critical evaluation of AI tools?
• How much exposure would I receive to consultation, procedures and multidisciplinary care?
• Am I comfortable with a career in which technology and workflows will keep changing?

The strongest reason to enter radiology is a good match with the work and willingness to keep learning—not confidence that the specialty will remain unchanged.

Which Skills Will Matter Most?

1. Clinical judgment: Connecting imaging findings with the actual clinical question.
2. Critical AI evaluation: Understanding false positives, false negatives and whether validation applies to the local patient population.
3. Handling uncertainty: Recognizing when a case does not fit an expected pattern and needs escalation.
4. Communication: Explaining findings and limitations to patients and other clinicians.
5. Quality improvement: Checking whether a new workflow improves care rather than simply generating more output.

Procedural expertise can add another dimension for radiologists whose practice includes interventions. These are practical development priorities, not guarantees against every future employment risk.

Frequently Asked Questions

Will AI replace radiologists completely?

Current evidence does not show that AI can replace the full range of radiologists' responsibilities across routine practice. It can automate selected tasks and may reduce human involvement in some validated workflows. Complete occupational replacement remains uncertain.

Can AI read scans better than a radiologist?

Some systems perform very well on specific detection or classification tasks. That does not establish superiority across all scans, diseases and clinical situations. Check whether a study tested AI alone, clinicians alone or an AI-supported workflow—and what outcome it measured.

Are radiologists and radiographers the same?

No. Radiologists are physicians who interpret imaging and may perform image-guided treatments. Radiographers, also called radiologic technologists in some countries, generally acquire images and work directly with patients and imaging equipment. Automation affects their tasks differently.

Which three jobs will survive AI?

No research can guarantee that exactly three occupations are permanently safe. Three examples with substantial barriers to full automation are electricians, bedside nurses and interventional radiologists. Their work involves physical environments, patient or client interaction, and handling situations that vary from case to case. AI can still change parts of each job.

Which five jobs will survive AI?

Five illustrative examples are electricians, plumbers, bedside nurses, physical therapists and interventional radiologists. This is a task-based assessment, not an official ranking or promise of job security. Each combines responsibilities that go beyond generating information on a screen.

What jobs will be gone by 2030?

There is no dependable list of occupations guaranteed to disappear by 2030. Routine clerical work is exposed to automation, but exposure is not the same as elimination. The ILO's 2025 assessment identifies job transformation as more likely than complete redundancy for most exposed occupations. Its analysis concerns generative AI, not every form of robotics or medical-image AI.

What jobs will no longer exist in five years?

Predicting the complete disappearance of an occupation within five years is usually more confident than the evidence allows. Employers may reduce particular positions or combine duties while the occupation continues elsewhere. Ask which tasks are becoming cheaper to automate and whether actual hiring data shows a change.

What career will never be replaced by AI?

No career can honestly be guaranteed never to change or face automation. Work involving unpredictable physical situations, personal relationships, judgment and responsibility presents additional barriers to full replacement. Those features offer reasons for resilience, not permanent immunity.

The Practical Conclusion

Radiology is highly exposed to AI because many of its tasks are digital. That does not make the whole profession easy to automate.

The evidence supports preparing for substantial changes in reading, reporting and workflow. It also supports taking the remaining clinical, procedural and organizational responsibilities seriously.

For someone considering radiology, the useful question is not “Can I find a career AI will never touch?” It is “Do I want this work, and am I prepared to develop the judgment and skills needed as it changes?”

Sources and Methodology

This article distinguishes clinical research, regulatory information, professional commentary and editorial scenarios. It does not assign a numerical probability of job replacement or claim that any career is permanently protected.

The mammography findings concern one studied workflow. Professional proposals are identified as proposals. Career examples are based on task characteristics, not a validated ranking of which jobs will survive.

• MASAI trial: Screening performance and characteristics of breast cancer detected with AI-supported mammography, 2025
• FDA: Artificial Intelligence-Enabled Medical Devices
• Radiology: Navigating the AI Landscape in Medical Imaging, 2025
• RSNA: Researchers Advocate for Separate Roles Between AI and Humans, July 2025
• RadiologyInfo: Interventional Radiology
• RadiologyInfo: Diagnostic Radiology Professions
• International Labour Organization: Generative AI and Jobs—A 2025 Update

Continue Reading

• Which Medical Specialties Are Safest From AI?
• AI in Radiology: Pros and Cons
• Will AI Replace Doctors?
• AI in Healthcare: Uses, Risks and What Comes Next
• What Jobs Will Get Replaced by AI?

Is ChatGPT Getting Worse? What the Data and Users Show

ChatGPT can get better on benchmarks and worse at a task you rely on. Both can be true. A newer model might solve harder coding problems while producing writing you dislike, overlooking an instruction or taking more corrections to finish familiar work. The evidence does not support a simple, continuous decline. It does show that updates can introduce real regressions—and that stronger overall scores do not guarantee a better everyday experience.

Updated September 20, 2026. This article separates independent evaluations, OpenAI's own reports and user-experience complaints.

Quick answer: ChatGPT is improving on some measured capabilities, but it is not becoming uniformly better for every user. Judge an update by factual accuracy, instruction following and the amount of correction your actual work requires. There is no single reliable percentage describing how often all ChatGPT answers are wrong.

Table of Contents

• Is ChatGPT Getting Better or Worse?
• What the Data Actually Shows
• How Often Is ChatGPT Wrong?
• Why ChatGPT May Feel Worse
• Common Complaints and What to Check
• Writing, Coding and Research: Judge Them Separately
• How to Test Whether It Has Gotten Worse for You
• How to Get Better Answers
• When to Try Another Tool
• The Verdict
• Frequently Asked Questions
• Sources and Methodology

Is ChatGPT Getting Better or Worse?

The answer depends on what you measure. Solving a difficult problem, reliably following a brief and writing in a voice you enjoy are different abilities. An update can improve one without improving the others.

Imagine that your old workflow produced a usable newsletter after one edit. A newer model scores higher on a coding test but now needs five corrections to stop using stock phrases. For your newsletter, that is a practical decline. It does not establish that the model has become worse at coding or mathematics.

The reverse also matters: one bad answer does not prove a general regression. To establish a trend, compare repeated attempts on similar tasks under comparable conditions.

What the Data Actually Shows

September 2026: Measurable Gains, With Some Trade-offs

In its September 9 evaluation, Artificial Analysis reported that GPT-6 Astra gained six points over GPT-5.6 Sol on its Intelligence Index. But it also reported a roughly 45-point Elo decline on GDPval-AA v2, an evaluation of professional tasks, and lower presentation-quality results in AA-Briefcase.

These are results from the configurations and evaluation environments tested, not a score for every ChatGPT conversation. They support a more useful conclusion than either “everything improved” or “everything got worse”: improvements can coexist with weaknesses on particular tasks.

OpenAI's Factuality Tests Measure a Different Question

The GPT-5.6 system card reports slightly fewer factual errors for Sol than GPT-5.5, with a larger reduction in reproducing specific errors users had flagged. OpenAI explicitly says these examples were selected for being error-prone. They are not a representative sample of everyday conversations.

That distinction prevents an easy mistake: a result on difficult, previously failed questions cannot tell you the overall likelihood that an ordinary answer will be wrong.

A Bad Behavioral Update Has Been Documented

In April 2025, OpenAI rolled back a GPT-4o update after it made responses excessively agreeable and flattering. This behavior is called sycophancy. The episode establishes that an update can make the product less useful even when it was intended to improve the experience.

It also gives user reports an appropriate role: complaints can reveal a specific failure worth investigating. Their existence alone does not establish how common that failure is across all users.

The Famous Stanford/Berkeley Study Is Historical Evidence

A 2023 study by Lingjiao Chen, Matei Zaharia and James Zou compared March and June versions of GPT-3.5 and GPT-4. It found changes in instruction following, code formatting and other tasks, with improvements in some areas and declines in others.

This demonstrates why ongoing evaluation matters. It does not measure the quality of ChatGPT in September 2026. Presenting a three-year-old result as proof of a current collapse would be misleading.

How Often Is ChatGPT Wrong?

The sources reviewed here do not establish a single error rate for all ChatGPT use. A useful percentage must identify the model, task, tools, settings and grading method. It must also explain what counts as an error: one incorrect detail, a wholly wrong answer, an unsupported citation or failure to follow instructions.

Checking a supplied invoice is different from recalling an obscure historical fact. A sourced answer can still misread its source. A polished paragraph can contain one consequential mistake. Combining these into one accuracy number hides the distinctions a reader needs.

What Did the “86% Hallucination Rate” Actually Mean?

Artificial Analysis's April 2026 GPT-5.5 evaluation reported 57% accuracy and an 86% hallucination rate for its xhigh configuration on AA-Omniscience. Those percentages use different denominators.

The benchmark's published methodology defines accuracy across all questions. Its hallucination rate measures incorrect answers among responses that were not correct, including partial answers and questions the model did not attempt:

Hallucination rate = incorrect ÷ (incorrect + partial answers + not attempted).

For illustration, if a model answers 60 of 100 questions correctly, answers 30 incorrectly and declines 10, its accuracy is 60%, but this hallucination rate is 75%: 30 divided by 40. This is a hypothetical example, not a ChatGPT test result.

So the reported 86% did not mean that 86% of ordinary ChatGPT answers were false. It highlighted a tendency to give incorrect answers rather than withhold an answer when the model could not answer correctly.

In September, Artificial Analysis reported a decrease in this metric from 92% for GPT-5.6 Sol to 51% for GPT-6 Astra at max effort, alongside improved accuracy. That is encouraging within this test. It still does not supply an everyday ChatGPT error rate.

For more on fabricated claims and references, see What Is a Hallucination in AI?

Why ChatGPT May Feel Worse

The Model or Experience You Preferred Changed

ChatGPT is a changing service. OpenAI's retirement announcement set February 13, 2026 as the removal date for GPT-4o and several older models from ChatGPT. It acknowledged that some users preferred GPT-4o's conversational warmth and creative ideation.

A preference for that style is not a misunderstanding of benchmarks. Tone, pacing and creative choices are part of the product's usefulness.

More Capable Does Not Necessarily Mean Easier to Direct

OpenAI's GPT-6 Astra developer guidance describes behavior that can create friction: asking questions when users expect the model to proceed, sensitivity to instructions in supporting files, and recurring phrasing. This is model guidance for developers, not proof that every ChatGPT user experiences these problems.

It illustrates why “intelligence” is too broad a diagnosis. The failure might concern initiative, competing instructions or style rather than inability to understand the subject.

Your Conversation May Contain Competing Directions

Consider a chat where you first requested a short summary, later asked for a detailed report and then returned to an earlier draft. If the answer keeps reverting to the short version, a clean conversation containing only the current brief is a useful diagnostic test.

This is a possible explanation, not an excuse for ignoring a clear correction. If the failure persists with a short, unambiguous prompt, record it as an instruction-following problem.

Your Standard for a Useful Answer May Have Changed

You may now notice errors you missed when the tool was new, or ask it to do more demanding work. That does not invalidate frustration. It means a fair comparison should use the same task, not compare today's difficult project with a memorable easy answer from last year.

Common ChatGPT Complaints and What to Check

The following are troubleshooting categories, not a survey of how frequently users encounter each problem.

Complaint	Useful check	What it tells you
It ignores my corrections.	Try the same task in a new chat with three explicit requirements.	A repeated failure is stronger evidence than one confused conversation.
It repeats itself or sounds generic.	Provide a short style example and identify the exact unwanted pattern.	You can separate a style mismatch from missing knowledge.
Answers are too short.	Specify required sections, examples and completeness.	Check whether it supplies the requested substance, not just more words.
It confidently invents facts.	Open the cited source and locate support for the claim.	A real link does not guarantee the claim is supported.
It agrees with a false premise.	Ask it to assess the premise against a supplied source.	Agreement is not independent verification.
It refuses a legitimate task.	Clarify the purpose and exact scope without changing the underlying task.	A refusal and an incorrect answer are different failure types.
It gets worse halfway through a project.	Restart from a concise record of current decisions.	This tests whether accumulated context contributes to the problem.

Writing, Coding and Research: Judge Them Separately

Writing and Editing

Evaluate whether the response preserves your meaning, follows your voice and respects the brief. A fluent rewrite that removes a crucial qualification is not an improvement. Save examples of writing you liked so you can compare concrete outputs rather than impressions.

Coding

Use working behavior as the test: does the change solve the problem, fit the existing code and pass relevant checks? A confident explanation of untested code should not count as success. Compare similar tasks in the same environment before concluding that coding performance has broadly declined.

Research

Check whether each important source exists, supports the attached claim and is current enough for the question. Ask for facts and interpretations to be separated. Neither a long reference list nor another chatbot's agreement independently validates an answer.

Everyday Conversation

A conversational tool can become less enjoyable without becoming worse at formal reasoning. If warmth, brevity or imaginative discussion is your main use, include that preference in your evaluation rather than treating it as irrelevant.

How to Test Whether ChatGPT Has Gotten Worse for You

Try this small comparison before repeatedly changing prompts or abandoning a workflow:

1. Choose five to ten representative tasks. Use work you actually repeat, such as editing a paragraph, extracting information or fixing a small bug.
2. Define success first. Write down the facts, format and constraints each answer must satisfy.
3. Keep conditions comparable. Record the date, visible model or mode, available tools, source files and relevant custom instructions.
4. Repeat each task in fresh chats. Three attempts can expose variability, although this remains a small personal test.
5. Track correction effort. Count factual errors, missed requirements, follow-up prompts and minutes spent repairing the output.
6. Compare with saved outputs or another available model. If you cannot access the old model or its original inputs, acknowledge that limitation.

This will not establish a worldwide trend. It can establish whether the tool is saving you less time on your own work—the question that matters when deciding how to use it.

How to Get Better Answers From ChatGPT

Start with a brief that describes the finished result. Include the audience, source material, non-negotiable requirements and an example where style matters. Remove old directions that no longer apply.

Try this prompt:

Complete the task using the material below. Follow these requirements: [list]. Preserve these facts: [list]. If a claim needs outside verification, identify it rather than inventing support. Make reasonable assumptions for minor choices and state them briefly. Ask a question only if the missing information would materially change the result. Before finishing, check the answer against the requirements.

For long work, use manageable stages such as source review, draft and final checks. For current factual questions, request current sources and inspect the important ones yourself. If the conversation keeps circling an abandoned approach, move the current brief into a fresh chat.

A self-check may catch omissions, but it is not independent fact-checking. Asking the same model whether it is certain is also not proof. Better instructions help expose failures; they do not remove the model's responsibility to follow a clear request or eliminate its limitations.

When to Try Another AI Tool

Try a second tool when the same well-defined task repeatedly fails, when you need a different writing style or when the correction work outweighs the time saved. Give both tools the same materials and success criteria.

Choose the output that actually meets your requirements, then verify important claims against the underlying sources. For factual disagreements, the deciding evidence should be the documents, calculations or tests—not a vote between chatbots.

The Verdict: Capability and Reliability Are Different Questions

The evidence supports neither a universal collapse nor a promise that every update is an upgrade for every user. Documented regressions exist, and some newer evaluations show gains alongside declines.

For a reader asking “Is ChatGPT getting worse?”, the useful test is concrete: Does it complete your work correctly, follow your instructions and require less repair? If those results deteriorate consistently, your workflow has become worse even if a headline benchmark improves.

Frequently Asked Questions

Is ChatGPT getting worse in 2026?

Not uniformly. The evaluations reviewed here show improvements in some capabilities and declines in others. A particular workflow can become less reliable without demonstrating that every use of ChatGPT has deteriorated.

Is ChatGPT getting better or worse at answering questions?

Specify the kind of question. A model's score on difficult knowledge tests does not establish its accuracy when summarizing your file or researching today's news. Judge the relevant task and check the supporting evidence.

How often is ChatGPT wrong?

There is no single rate established by these sources for all ChatGPT conversations. Error rates depend on the model, task, tools and definition of an error. Always check what a benchmark percentage actually measures.

Why does ChatGPT ignore instructions and repeat itself?

Possible contributors include competing directions, accumulated conversation context and model behavior. Try a fresh chat with a short, specific brief. If the same failure persists, treat it as a reproducible problem rather than assuming you need an increasingly elaborate prompt.

Does a confident answer mean ChatGPT checked the facts?

No. Confident wording is not evidence that a claim was verified. Ask for the supporting source, open it and check whether it actually supports the answer.

Should I pay for ChatGPT if the answers feel worse?

Base the decision on your own repeated tasks. Compare usable results and correction time with the tools you already have. A subscription is valuable to you only if the access and features you use justify the cost.

Sources and Methodology

This is an evidence review, not an original benchmark or a representative survey of ChatGPT users. Numerical results are attributed to the organizations that published them. User-experience examples and troubleshooting suggestions are illustrative, not estimates of how often problems occur.

Evaluations can differ in model settings, tools, task selection and grading. Compare models within the same reported evaluation; do not combine scores from different versions of a benchmark into a single trend line. API and coding-agent results should not be treated as direct measurements of every ChatGPT mode.

• Artificial Analysis: Benchmarking GPT-6 Astra, September 9, 2026
• Artificial Analysis: GPT-5.5 evaluation, April 23, 2026
• Artificial Analysis: AA-Omniscience definitions and methodology
• OpenAI: GPT-5.6 system card, factuality evaluation
• OpenAI: Sycophancy in GPT-4o
• OpenAI: Retirement of GPT-4o and older ChatGPT models
• OpenAI developer guidance: GPT-6 Astra behavior
• Chen, Zaharia and Zou: How Is ChatGPT's Behavior Changing Over Time? Revised October 2023

Continue Reading

• What Is a Hallucination in AI?
• Basics of Artificial Intelligence
• AI Ethics: Core Principles and Real Risks

Will AI Replace Cooking at Home

AI can already write a recipe, plan your meals and tell you what to make from ingredients in your refrigerator. But that isn't replacing cooking. The real breakthrough comes when you can say, "Make chicken curry and rice for four at 7 PM," and a machine finds the ingredients, prepares them, cooks the meal and cleans up afterward.

Robots are beginning to perform surprisingly complex household tasks, but a truly autonomous robot cook still has some very difficult problems to solve.

Short answer: AI is likely to automate more of home cooking, but today's general-purpose home robots cannot reliably replace a human cook in an ordinary kitchen. Specialized cooking machines can automate individual recipes or parts of meal preparation. The much bigger breakthrough will be a dexterous household robot that can use the kitchen you already have.

Table of Contents

• What Would Actually Count as Replacing Home Cooking?
• What Can AI Already Do in the Kitchen?
• Can Today's Home Robots Actually Cook?
• Why Humanoid Robots Could Change Everything
• Why Cooking Is So Difficult for a Robot
• Can a Robot Chop Vegetables?
• What About Raw Meat and Food Safety?
• Can AI Taste Food?
• The Forgotten Problem: Who Cleans Up?
• Robot Kitchen vs Humanoid Robot
• How Much Could a Robot Cook Cost?
• When Could Robots Really Replace Home Cooking?
• What Parts of Cooking Will Disappear First?
• Will People Still Cook?
• Could Robot Cooking Be Transformative for Older and Disabled People?
• Bottom Line
• Frequently Asked Questions

What Would Actually Count as Replacing Home Cooking?

We need a stricter definition than many "AI cooking" demonstrations use.

If ChatGPT gives you a recipe and you spend 45 minutes chopping and cooking it, AI hasn't replaced cooking.

If a smart oven automatically selects the temperature, AI hasn't replaced cooking.

If a countertop appliance stirs food after you wash, peel, cut and measure every ingredient, it has automated part of cooking—but you are still doing much of the work.

For AI to genuinely replace everyday home cooking, a system would need to perform most of the physical workflow.

Imagine saying:

"Make chicken curry, rice and vegetables for four. Dinner at 7."

A genuine robotic cook would ideally be able to:

1. Understand what you want.
2. Know which ingredients are available.
3. Retrieve them from cabinets and the refrigerator.
4. Wash ingredients when necessary.
5. Peel and cut vegetables.
6. Safely handle raw meat.
7. Measure ingredients.
8. Use pots, pans, knives and appliances.
9. Adjust heat and cooking time.
10. Recognize when food is properly cooked.
11. Serve the meal.
12. Store leftovers.
13. Clean the cookware and work surfaces.

That is an enormously more difficult robotics problem than generating a recipe.

Our test: If you still have to prep all the ingredients, move everything into the machine and clean up afterward, the robot hasn't replaced home cooking. It has made cooking easier.

What Can AI Already Do in the Kitchen?

The thinking side of cooking is becoming automated much faster than the physical side.

Today's AI systems can already help with:

• Generating recipes.
• Planning weekly meals.
• Adapting recipes to dietary restrictions.
• Substituting missing ingredients.
• Creating shopping lists.
• Scaling recipes for different numbers of people.
• Explaining cooking techniques.
• Estimating cooking times.
• Suggesting meals from available ingredients.

A multimodal AI can potentially look at ingredients and reason about what could be made from them.

1X, for example, says its NEO home robot combines visual intelligence with a built-in language model and can recognize ingredients on a kitchen counter and suggest dishes.

That's useful—but notice the distinction.

Knowing what to cook is much easier than physically cooking it.

Can Today's Home Robots Actually Cook?

Not in the way most people mean when they imagine a robotic housekeeper.

There are specialized automated cooking appliances and experimental robotic kitchens capable of impressive food preparation. But they generally operate within much more controlled conditions than a human cooking in an ordinary household kitchen.

The more interesting development is the arrival of general-purpose home robots.

1X is taking orders for NEO, a humanoid robot designed specifically for household work. The company lists an Early Access purchase price of $20,000 and a $499-per-month subscription option, with U.S. deliveries beginning in 2026.

NEO has hands, arms, vision, mobility and an AI system designed to learn household chores.

But there is a very important caveat.

1X says early NEO units arrive with basic autonomy. For complex chores the robot doesn't yet know, an expert can remotely supervise or guide the robot so it can complete and learn the task.

This distinction matters: A robot performing a kitchen task while a remote human helps control or teach it is evidence that the hardware can perform the movements. It is not evidence that a fully autonomous robot can independently cook dinner every night.

Why Humanoid Robots Could Change Everything

Our homes weren't designed for robots.

They were designed for us.

Cabinet handles are positioned for human hands. Countertops are built at human working height. Knives, refrigerators, faucets, dishwashers and stove controls all assume a roughly human body.

That gives humanoid robots an interesting advantage.

Instead of rebuilding the kitchen around a machine, manufacturers are trying to build a machine that can operate a human kitchen.

Figure demonstrated an important step in January 2026 with its Helix 02 AI system.

A humanoid robot autonomously unloaded and reloaded a dishwasher across a full-sized kitchen during a continuous four-minute task. Figure says the demonstration required the robot to integrate vision, walking, balance and manipulation without human intervention or resets.

That isn't cooking.

But it demonstrates several abilities that a future robot cook will need: moving through a kitchen, grasping fragile objects, opening drawers and manipulating ordinary household equipment.

The difference between "robot can load the dishwasher" and "robot can prepare tonight's dinner" is still enormous.

But the direction is important.

Why Cooking Is So Difficult for a Robot

Humans underestimate cooking because we've practiced manipulating objects our entire lives.

Consider something as simple as making an omelet.

A robot might have to:

• Find the eggs.
• Pick one up without crushing it.
• Crack the shell without dropping pieces into the bowl.
• Recognize whether the egg is spoiled.
• Whisk several eggs.
• Find an appropriate pan.
• Add the right amount of oil or butter.
• Control the stove.
• Recognize when the pan is hot.
• Pour the eggs without spilling them.
• Judge when they are cooked.
• Fold or turn the omelet.
• Transfer it onto a plate.

And that's a relatively simple meal.

Now imagine a whole chicken, slippery vegetables, boiling water, splattering oil and three dishes cooking simultaneously.

A language model can explain those steps almost instantly.

A physical robot has to execute every one of them without injuring someone, breaking something, contaminating the food or starting a fire.

Can a Robot Chop Vegetables?

Robots can manipulate and cut food under controlled conditions, but reliably chopping arbitrary ingredients in an ordinary kitchen is much harder.

A tomato behaves differently from a potato.

An onion rolls.

A carrot is hard.

A ripe mango can be slippery.

Vegetables vary in size and shape every time.

Humans continuously adjust grip pressure, knife angle, cutting force and finger position without consciously thinking about it.

For a home robot, knife use also creates an obvious safety problem.

The robot doesn't merely need enough dexterity to use a knife. It needs sufficient perception and control to use a sharp blade safely around children, pets and humans moving unpredictably through the kitchen.

This is one reason early robot cooking may depend more heavily on pre-cut ingredients, specialized appliances or safer preparation methods.

What About Raw Meat and Food Safety?

Cooking isn't just a manipulation problem.

It is a sanitation problem.

Imagine a robot handling raw chicken and then reaching for a refrigerator handle.

A human cook knows—or should know—that the hand needs washing first.

A reliable robot cook must understand cross-contamination and maintain hygienic workflows.

It may need to distinguish:

• Raw from cooked food.
• Clean from contaminated utensils.
• Safe internal temperatures.
• Food that has been left unrefrigerated too long.
• Allergens.
• Spoiled ingredients.

These aren't minor details.

A robot that occasionally folds a shirt incorrectly is annoying.

A robot that occasionally undercooks chicken can make someone sick.

Reliability requirements change with the task. A home robot doesn't have to be perfect at folding towels. A robot handling knives, flames and food safety has to meet a much higher standard.

Can AI Taste Food?

This is another surprisingly difficult part of replacing a human cook.

Humans don't cook purely from a timer.

We look, smell and taste.

We notice that the sauce needs salt, the onions aren't browned enough, the dough feels too wet or the curry is becoming too thick.

A robot can use cameras, temperature sensors, scales and other instruments to measure aspects of cooking more precisely than a human.

Specialized automated cooking systems are also being developed around sensor-based monitoring.

But reproducing the combined human senses of taste, smell, texture and experience across thousands of different dishes remains much more complicated.

The solution may not require a robot to taste exactly like a person.

Instead, future cooking systems could combine sensors with recipes and feedback from household members:

"Less salt next time."

"Cook the vegetables a little longer."

"Make this curry hotter than last time."

An AI system with memory could then personalize meals over time.

The Forgotten Problem: Who Cleans Up?

This may determine whether consumers consider robot cooking genuinely useful.

Imagine paying $20,000 for a robot that cooks dinner but leaves you with:

• Three dirty pans.
• A cutting board covered with food.
• Spilled ingredients.
• A greasy stovetop.
• Dirty knives.
• Food scraps on the counter.

You haven't eliminated kitchen work.

You have changed which kitchen work you do.

This is why recent progress in apparently mundane chores such as dishwasher loading is important.

A robot that can cook but cannot clean has solved only part of the problem.

The real home-cooking breakthrough is likely to require an integrated loop:

Get ingredients → prepare → cook → serve → store leftovers → clean.

Robot Kitchen vs Humanoid Robot: Which Will Win?

There are two very different ways to automate home cooking.

Dedicated Robotic Kitchen	Humanoid Home Robot
Kitchen designed around automation	Robot designed around existing kitchens
Can optimize equipment for robotic use	Can potentially use ordinary appliances
Potentially very reliable at supported recipes	Potentially much more flexible
May require major kitchen installation	Could work in multiple rooms
Main purpose is cooking	Could cook, clean, do laundry and perform other chores
Large fixed investment	One robot potentially replaces several specialized machines

Specialized robotic kitchens may achieve high-quality automated cooking sooner because the environment can be controlled.

But the humanoid model could ultimately be much more disruptive.

If you already own a refrigerator, oven, dishwasher, microwave and stove, why buy an entirely new robotic kitchen if one general-purpose robot can eventually use all of them?

That is the long-term promise behind humanoid household robots.

How Much Could a Robot Cook Cost?

Today's pricing shows just how early this market remains.

1X lists NEO Early Access ownership at $20,000, while its subscription is listed at $499 per month.

And today's NEO is not being sold as a fully autonomous personal chef capable of replacing all home cooking.

For most households, those economics don't yet make sense purely for cooking.

But consider what happens if a future robot can perform several hours of household labor every day:

• Cooking.
• Dishwashing.
• Laundry.
• General tidying.
• Fetching objects.
• Taking out trash.
• Helping an older family member.

Then consumers aren't comparing its cost with a blender or air fryer.

They're comparing it with the value of hundreds or thousands of hours of household labor.

That could completely change the economic calculation.

When Could Robots Really Replace Home Cooking?

No one can responsibly give an exact date.

Robotics progress can accelerate rapidly in one area while remaining stubbornly difficult in another.

A reasonable way to think about the future is in stages rather than promises.

Stage	Likely Capability
Now	AI plans meals and recipes; specialized appliances automate individual cooking processes; advanced robots demonstrate household manipulation tasks.
Next several years	Home robots could become useful assistants—fetching ingredients, loading appliances, cleaning and performing selected food-preparation tasks.
Later	Robots may prepare a limited library of complete meals in suitably arranged kitchens with minimal human assistance.
Longer term	A general-purpose robot could potentially enter an ordinary kitchen, select ingredients and independently prepare a wide variety of meals from beginning to end.

The last stage is what would truly qualify as replacing everyday home cooking.

I would be skeptical of anyone confidently assigning an exact year to it.

We have evidence that robots are gaining the physical abilities needed for household work. We do not yet have evidence that a mass-market robot can safely and reliably cook arbitrary meals in millions of unpredictable home kitchens.

What Parts of Cooking Will Disappear First?

Full automation isn't necessary for AI and robotics to substantially reduce kitchen work.

The first tasks to become commonplace may be the repetitive ones:

• Meal planning.
• Grocery inventory.
• Shopping lists and ordering.
• Measuring ingredients.
• Monitoring temperature.
• Stirring.
• Timing.
• Dishwashing.
• Basic cleanup.
• Preparing standardized meals.

Humans may continue handling unusual dishes and creative cooking while machines take over the repetitive weekday work.

In other words, AI might reduce home cooking dramatically before it completely replaces it.

Will People Still Cook Even If Robots Can Do It?

Almost certainly.

Washing clothes by hand became unnecessary for many households after washing machines arrived. People generally didn't mourn the lost chore.

Cooking is different.

For some people it is labor. For others it is:

• A hobby.
• A family tradition.
• A cultural activity.
• A creative outlet.
• A way of caring for other people.

So a robot capable of cooking doesn't mean humans stop cooking.

It means cooking could become increasingly optional.

You might cook biryani yourself on Sunday because you enjoy it, then tell the robot to make dinner Monday through Thursday because you're tired after work.

That is probably a better analogy than "robots eliminate cooking."

The biggest change may be that cooking becomes a choice rather than a daily obligation. People who love cooking can continue doing it. People who don't may eventually delegate much of it to a machine.

Could Robot Cooking Be Transformative for Older and Disabled People?

This may ultimately be more important than saving busy professionals 45 minutes at night.

Cooking can become difficult for someone who has limited mobility, reduced hand strength or difficulty standing for long periods.

A robot capable of retrieving food, safely preparing meals and cleaning afterward could help some people remain independent at home longer.

The same general-purpose robot might also retrieve objects, load a dishwasher and perform other physical household tasks.

Airational has previously examined this broader possibility in The Future of Robotic Aides for the Elderly.

But reliability and safety become even more important when a person depends on the machine rather than merely using it for convenience.

Bottom Line: Will AI Replace Cooking at Home?

Probably some day for people who want it to—but we're not there yet.

AI has already become surprisingly capable at the intellectual side of cooking: recipes, planning, substitutions and instructions.

The physical side remains the bottleneck.

A truly useful robot chef needs dexterity, mobility, vision, memory, safety awareness and enough general intelligence to handle a kitchen where ingredients, utensils and people are never in exactly the same place twice.

Recent humanoid-robot progress makes that future more believable. Robots can already demonstrate increasingly complex household manipulation, and the first general-purpose humanoid home robots are moving toward consumers.

But demonstrations should not be confused with a finished autonomous cook.

The test is simple:

Tell the robot, "Make dinner."

If you have to find the ingredients, chop everything, load the machine and clean the kitchen afterward, AI hasn't replaced cooking.

If you can leave the room and return to a cooked meal and a clean kitchen, it has.

That second future still requires major advances in robotics.

But unlike the idea of an AI recipe generator, it would represent something genuinely transformative: AI moving from telling humans how to do household work to physically doing the work for them.

Explore More About AI and Automation

• The Future of Robotic Aides for the Elderly
• The Future of AI in Manufacturing
• AI Drive-Thru Revolution
• What Jobs Will Get Replaced by AI?
• Basics of Artificial Intelligence

Frequently Asked Questions

Can AI cook food by itself?

Specialized automated cooking systems can prepare certain foods with limited human intervention, and experimental robots can perform increasingly complicated kitchen tasks. However, today's general-purpose consumer robots cannot reliably prepare arbitrary meals from beginning to end in an ordinary home kitchen.

Can a humanoid robot cook dinner?

Humanoid robots have demonstrated individual household and kitchen-related tasks, but there is a large difference between demonstrations and reliably cooking a complete dinner every day. General-purpose autonomous cooking remains an unsolved consumer robotics challenge.

Can a robot chop vegetables?

Robotic systems can cut and manipulate food under controlled conditions. Doing it safely and reliably with vegetables of different shapes, sizes, textures and positions in an ordinary kitchen is considerably harder.

Will robots be able to use our existing kitchens?

That is one of the major potential advantages of humanoid robots. Their human-like shape and hands could allow them to operate appliances and tools designed for people rather than requiring homeowners to install an entirely robotic kitchen.

Will AI be able to make Indian food such as curry or biryani?

There is nothing inherently impossible about a robot preparing complex cuisines. The challenge is physical execution: measuring spices, cutting ingredients, controlling heat, judging consistency, handling multiple cooking stages and cleaning afterward. A sufficiently capable general-purpose robot could eventually learn these workflows.

Could I tell a future robot what I want for dinner?

That is a plausible long-term interface. Language models already understand requests such as meal preferences and dietary restrictions. The difficult part is connecting that intelligence to a physical robot capable of safely performing all the necessary kitchen actions.

How much does a home humanoid robot cost?

Prices remain extremely high for early products. For example, 1X lists its NEO Early Access robot at $20,000 or a $499-per-month subscription. Those prices and capabilities are likely to change substantially as the market develops.

How soon will robots cook all our meals?

There is no reliable date. Limited cooking assistance is likely to arrive much sooner than complete autonomous meal preparation. Replacing a human across arbitrary recipes and ordinary kitchens requires much greater reliability, dexterity and safety than today's consumer home robots have demonstrated.

Will AI make home cooking obsolete?

Probably not. Even if robots eventually make cooking optional, many people cook for enjoyment, culture and family traditions. Automation is more likely to remove unwanted routine cooking than eliminate cooking as a human activity.

Will Robotaxis Replace Uber Drivers? How Soon Could It Really Happen?

Could Uber drivers become unnecessary in some cities within the next few years? Robotaxis are no longer just prototypes. Driverless and supervised autonomous rides are already operating in multiple markets, Tesla has started producing its purpose-built Cybercab, and Uber is building partnerships designed to put autonomous vehicles directly into the same app that millions of human drivers use today. The transition probably won't make Uber drivers disappear everywhere at once—but the first economic effects could arrive long before the last human driver leaves the road.

Short answer: Robotaxis are unlikely to make all Uber drivers extinct within a few years. But drivers in dense, robotaxi-friendly cities could face fewer available rides and pressure on earnings much sooner. The real disruption begins when autonomous cars become cheaper and plentiful enough to compete with human drivers for ordinary Uber trips.

Table of Contents

• Robotaxis Are Already Taking Real Rides
• Uber's Strategy: Don't Fight Robotaxis—Put Them on Uber
• What Tesla Cybercab Changes
• The Economics: Human Uber vs Robotaxi
• Do Robotaxis Actually Hurt Driver Income?
• Which Uber Drivers Could Be Affected First?
• Which Drivers Are Harder to Replace?
• Why Can't Robotaxis Replace Drivers Everywhere Yet?
• Uber Says Robotaxis Will Complement Drivers
• When Could Uber Drivers Actually Be Replaced?
• Robotaxis Remove Drivers—but Create Other Jobs
• Should Uber Drivers Be Worried?
• Bottom Line
• Frequently Asked Questions

Robotaxis Are Already Taking Real Rides

The biggest change in the robotaxi debate is simple: we no longer have to ask whether autonomous ride-hailing is technically possible.

It is happening.

Tesla says its Robotaxi service is operating with Model Y vehicles in Austin, Dallas, Houston, Miami and Tampa. Its purpose-built Cybercab is already carrying passengers in limited areas of Austin.

Uber is also putting autonomous vehicles from multiple partners directly into its ride-hailing network.

That means a passenger requesting an ordinary ride can increasingly be matched with a vehicle in which driving is performed partly or entirely by an autonomous system.

This changes the employment question.

The question is no longer:

"Will a computer ever be able to drive a taxi?"

It is becoming:

"How quickly can autonomous fleets become cheap, reliable and widespread enough to take a meaningful percentage of rides currently performed by humans?"

Uber's Strategy: Don't Fight Robotaxis—Put Them on Uber

Uber's position in the autonomous revolution is particularly interesting.

Uber doesn't need to manufacture every autonomous vehicle itself.

Instead, it can become the marketplace connecting passengers with vehicles supplied by autonomous-driving companies and fleet operators.

Uber says autonomous vehicles are already live on its platform in multiple cities and that it is working with more than 30 autonomous-vehicle partners across mobility, delivery and freight.

In its second-quarter 2026 investor remarks, Uber said partners had committed approximately 120,000 autonomous vehicles to the Uber network over the coming years.

The company also said it expects to commit more than $10 billion across investments, infrastructure and vehicle commitments as it tries to become the leading commercialization platform for autonomous mobility.

This is the paradox for Uber drivers: Uber says human drivers remain central to its marketplace, while simultaneously investing heavily in infrastructure designed to make enormous autonomous fleets commercially viable.

Uber's partnerships already include different autonomous-driving companies and automakers.

For example, Uber and Rivian announced plans for an initial 10,000 fully autonomous R2 robotaxis beginning in San Francisco and Miami in 2028, with an option that could eventually bring the arrangement to as many as 50,000 vehicles.

Uber and Zoox have also announced plans to bring purpose-built Zoox robotaxis onto the Uber platform, beginning in Las Vegas and later Los Angeles.

MOIA, part of Volkswagen Group, is testing autonomous ID. Buzz vehicles for Uber in Los Angeles with plans for commercial rides.

The strategy is becoming clear: Uber wants the Uber app to remain valuable whether the vehicle arriving has a human driver or an AI driver.

What Tesla Cybercab Changes

Tesla's Cybercab makes the employment implications unusually obvious because it is not designed as a normal car that happens to drive itself.

It is a purpose-built autonomous vehicle.

Tesla says Cybercab has no steering wheel and is designed to navigate autonomously.

The company began production of Cybercab in 2026 and started engineering test drives of production vehicles on public roads. Tesla has also begun offering Cybercab rides in limited areas of Austin.

A vehicle with no steering wheel isn't being designed around a future in which a rideshare driver sits behind it.

That doesn't mean Tesla can instantly deploy millions of Cybercabs.

But it illustrates where the technology is trying to go:

ride-hailing in which paying a human to drive is no longer part of the transaction.

The Economics: Human Uber vs Robotaxi

Technology alone won't decide whether robotaxis replace Uber drivers.

Economics will.

Consider a simplified ride.

Human-Driven Uber	Robotaxi
Vehicle cost/depreciation	Vehicle cost/depreciation
Fuel or charging	Charging
Insurance	Commercial/fleet insurance
Maintenance	Maintenance
Driver compensation	No onboard driver compensation
Driver handles vehicle between rides	Fleet operations required
Driver may clean vehicle	Fleet must clean vehicle
Driver handles many unexpected situations	Remote/fleet assistance may be required

The robotaxi eliminates one major expense—the human driver—but adds others.

Someone still has to finance the vehicle, charge it, clean it, maintain it, insure it, recover it when something goes wrong and keep the fleet operating.

That means robotaxis don't automatically win economically simply because they don't have drivers.

But autonomous vehicles have another potential advantage: utilization.

A human driver needs sleep, meals and personal time.

A fleet vehicle can theoretically operate for much longer periods each day, stopping mainly for charging, cleaning and maintenance.

If a robotaxi can complete enough paid rides per day while maintaining acceptable operating costs, its economics could eventually become difficult for a human driver to compete against.

The tipping point isn't "AI can drive." The real tipping point is when a robotaxi can reliably complete the same useful rides as a human driver at a lower total cost per mile.

Do Robotaxis Actually Hurt Driver Income?

We now have some real-world evidence that they can.

A 2026 study published in Humanities and Social Sciences Communications examined the introduction of Baidu's Apollo Go robotaxis in Wuhan, China.

Researchers analyzed more than 200,000 daily observations from traditional taxi drivers.

They found that the introduction of robotaxis was associated with a 10.9% decline in traditional taxi drivers' average daily income in the affected area.

The surveyed drivers also reported longer working hours, greater job stress, lower job satisfaction and increased interest in finding alternative employment.

This is an important distinction: Robotaxis don't have to eliminate every driver's job to hurt drivers economically. If autonomous vehicles take enough profitable rides, human drivers can experience lower earnings while they are still technically employed.

One study in one Chinese city cannot tell us exactly what will happen to Uber drivers in the United States.

Different regulations, labor markets, pricing structures and transportation patterns matter.

But it provides evidence that robotaxi competition can affect driver income before full job replacement occurs.

Which Uber Drivers Could Be Affected First?

The earliest pressure is likely to occur where autonomous vehicles work best.

That generally means:

• Dense metropolitan areas.
• Frequently traveled routes.
• Predictable road environments.
• Areas with strong passenger demand.
• Markets where autonomous operation has regulatory approval.
• Trips that fit the vehicle's passenger and luggage capacity.

A driver making most of their income from ordinary urban rides in a city with thousands of robotaxis could face more competition than a driver operating in an area where autonomous service remains unavailable.

Airport trips could eventually become especially important because they are often valuable rides, although airports can also impose complicated operational and regulatory requirements.

Which Drivers Are Harder to Replace?

Robotaxis are likely to expand unevenly.

Human drivers retain advantages in situations involving:

• Remote or rural destinations.
• Unusual pickup locations.
• Road closures and unpredictable construction.
• Severe weather.
• Passengers needing extra assistance.
• Large groups or unusual luggage.
• Special events with chaotic traffic patterns.
• Situations where human judgment or communication is useful.

This is one reason the transition could produce a hybrid market for years.

Robotaxis may capture the easiest and most repeatable trips first while human drivers continue serving the long tail of complicated trips.

Unfortunately for drivers, that could create another problem.

If autonomous fleets take many of the easy, efficient rides, human drivers could be left competing disproportionately for less convenient trips.

Why Can't Robotaxis Replace Drivers Everywhere Yet?

Driving in a carefully supported autonomous-service area is different from operating anywhere a passenger might request an Uber.

Robotaxis still face significant obstacles.

Regulation

Autonomous-driving rules vary dramatically among states, countries and cities.

Political opposition can also emerge when deployments expand.

In September 2026, Minneapolis City Council members proposed requiring autonomous vehicles to carry paid human safety monitors, with supporters raising both safety and employment concerns.

Weather

Heavy snow, flooding, fog and other difficult conditions can complicate autonomous driving and sensor performance.

Unpredictable Human Behavior

Roads contain pedestrians, cyclists, emergency vehicles, construction workers and human drivers who don't always follow rules.

Fleet Operations

Driverless cars don't clean or repair themselves.

Uber's Tokyo robotaxi plans provide a useful example. Its local fleet partner is expected to handle depot operations, cleaning, maintenance, inspections, charging and vehicle availability.

Capital

Human Uber drivers commonly supply the vehicle themselves.

A robotaxi network requires somebody else to finance potentially enormous fleets.

Uber's willingness to commit billions of dollars to autonomous mobility shows how capital-intensive that transition can be.

Uber Says Robotaxis Will Complement Drivers

Uber currently describes its strategy as a hybrid marketplace.

The company tells drivers that autonomous vehicles are designed to complement rather than replace them.

Uber argues that AVs remain limited geographically and can help serve demand during busy periods or where there aren't enough drivers.

That can certainly be true during the early deployment phase.

If a city has one million ride requests and only enough robotaxis to handle 10,000 of them, human drivers remain indispensable.

But the long-term employment question depends on what happens as that percentage increases.

If autonomous vehicles eventually handle 5% of rides, driver impact may be limited.

At 20%, competition becomes more meaningful.

At 50% or 70%, the economics of driving for Uber could look radically different even if human drivers technically remain on the platform.

Watch market share, not just job counts. The earliest warning for Uber drivers may not be an announcement that drivers are being eliminated. It could be a gradual decline in the percentage of rides available to humans.

When Could Uber Drivers Actually Be Replaced?

No credible source can give an exact year when Uber drivers will disappear.

The transition is likely to happen city by city rather than nationally.

Period	What Could Happen
2026–2028	Robotaxis expand in selected cities, but human drivers continue performing the overwhelming variety of rides in most markets.
2028–2032	If costs, safety and regulation continue improving, large autonomous fleets could begin materially competing with human drivers in major metropolitan markets.
2030s	Driver displacement could become much more significant if autonomous vehicles expand beyond selected urban zones and prove consistently cheaper than human-driven rides.
Complete replacement	Highly uncertain. Rural routes, unusual conditions, regulation and specialized passenger needs could preserve human driving much longer.

These are scenarios, not predictions.

One clue to the potential scale comes from Uber itself. Its Rivian agreement targets initial autonomous deployments in 2028 and expansion to 25 cities through 2031.

That makes the late 2020s and early 2030s particularly important to watch.

Robotaxis Remove Drivers—but Create Other Jobs

A driverless taxi doesn't mean a workerless taxi business.

Large autonomous fleets need people for:

• Vehicle cleaning.
• Charging.
• Maintenance and repairs.
• Fleet inspections.
• Depot operations.
• Customer support.
• Remote assistance.
• Software and hardware engineering.
• Mapping and data operations.
• Fleet management.

Uber's Tokyo arrangement demonstrates this clearly: a traditional taxi operator is being brought into the autonomous system to manage the physical fleet.

But there is no reason to assume one robotaxi creates one replacement job.

A single worker could potentially clean, charge, monitor or maintain many vehicles.

So employment may shift while the total amount and type of labor required changes.

Should Uber Drivers Be Worried?

If robotaxis aren't operating anywhere near you, the immediate effect may be negligible.

If you drive in an early autonomous market, the issue deserves much closer attention.

The metrics worth watching are:

• How many autonomous vehicles are operating locally.
• What percentage of Uber rides they perform.
• Whether autonomous service expands geographically.
• Whether robotaxis begin serving airports.
• Robotaxi fares compared with human-driven rides.
• Changes in driver wait times between trips.
• Changes in driver earnings per hour.
• Whether autonomous fleets begin operating during peak periods.

Those indicators tell drivers much more than futuristic predictions about when "all cars will drive themselves."

Bottom Line: Will Robotaxis Replace Uber Drivers?

Some Uber driving jobs are likely to be displaced if robotaxis continue becoming cheaper, more capable and more widely deployed.

That process has already moved beyond laboratory testing.

Tesla is producing Cybercab. Uber is integrating autonomous vehicles from numerous partners. Purpose-built robotaxis are entering commercial ride-hailing networks. And we now have real-world research showing robotaxi deployment can reduce traditional drivers' earnings.

But "Uber drivers will be extinct in a few years" goes beyond the evidence.

Robotaxis still face regulatory, geographic, technical and economic constraints. Human drivers can go almost anywhere a road and regulations permit; autonomous systems are still expanding market by market.

The most plausible transition is therefore not:

Human drivers today → zero human drivers tomorrow.

It is:

Human-dominated rideshare → hybrid fleets → autonomous dominance in some cities → continued human driving where autonomy remains difficult or uneconomical.

The real danger to Uber drivers isn't necessarily extinction. It's economics. Drivers can lose income, bargaining power and attractive rides long before the occupation completely disappears.

Explore More About AI and Jobs

• What Jobs Will Get Replaced by AI?
• Top 15 Jobs AI Will Replace
• AI Job Replacement Risk Calculator
• Why AI Hasn't Taken Your Job Yet—and What Comes Next
• The Future of Self-Driving Trucks
• AI Drive-Thru Revolution

Frequently Asked Questions

Will Uber drivers be extinct in a few years?

Probably not. Robotaxis are expanding quickly, but autonomous services still operate in limited markets and face regulatory, technical and economic constraints. Some cities could experience meaningful driver displacement much sooner than others.

Will Tesla Cybercab replace Uber drivers?

Cybercab is specifically designed for autonomous ride-hailing and has no steering wheel. If Tesla can deploy it economically at large scale, it could compete directly with human rideshare drivers. However, Cybercab deployment is still limited and large-scale expansion remains uncertain.

Does Uber want to replace its drivers with robotaxis?

Uber publicly says autonomous vehicles are intended to complement drivers through a hybrid marketplace. At the same time, Uber is investing heavily in autonomous mobility and has partnerships that could eventually put very large numbers of autonomous vehicles onto its network.

Are robotaxis already affecting taxi-driver income?

There is evidence that they can. A 2026 study of Baidu Apollo Go deployment in Wuhan found a 10.9% short-run decline in average daily income among traditional taxi drivers in the affected area. Results in other countries and rideshare markets may differ.

When will robotaxis become common?

They are already becoming common in selected service areas, but nationwide or global availability is much further away. The late 2020s and early 2030s could be an important expansion period if current deployment plans succeed.

Will Waymo replace Uber drivers?

Waymo and other autonomous-driving companies can reduce demand for human drivers wherever autonomous rides compete for the same passengers. The effect depends on fleet size, geographic coverage, pricing and passenger adoption.

What happens to Uber drivers when robotaxis arrive?

The first effect may be fewer available rides or lower earnings rather than immediate job elimination. Drivers could remain active on the platform while competing with autonomous vehicles for passenger demand.

What jobs will robotaxis create?

Autonomous fleets require maintenance, cleaning, charging, inspections, fleet operations, customer support, engineering and other services. However, there is no guarantee that the number of new jobs will equal the number of driving opportunities eventually displaced.

Employment note: Autonomous-vehicle technology, regulation and deployment plans are changing rapidly. Timelines discussed here are scenarios based on current developments, not guarantees about future employment or the date at which autonomous vehicles will reach a particular market.