FTC Chair Says AI Developers Could Be Liable for Agent Harm

What did the FTC chair say about AI agent liability?

FTC Chair Andrew Ferguson said AI developers could be liable when AI agents they instruct cause harm. Ferguson made the remarks on September 25, 2026, at the Reuters Momentum AI Austin event, where he argued that the central policy question is not what to do about the tool after it acts, but what responsibility belongs to the person or company directing it. Reuters reporting on Ferguson’s remarks provides the direct account of that position.

AI agent liability matters because newer AI systems can do more than produce a response in a chat window. An agent can be given instructions, access to connected services, and permission to complete tasks. Those capabilities can make a mistake, a deceptive interaction, or an unauthorized transaction more consequential than an incorrect answer generated by a basic chatbot.

Ferguson’s comments are not a new FTC rule or a final legal determination about any specific developer. The practical point is narrower and more immediate: businesses should not assume that calling a system an “agent” removes responsibility for how the system was designed, instructed, or deployed.

Why does AI agent liability focus on human and company instructions?

AI agent liability, under Ferguson’s framing, begins with the developer or business that gives an agent its instructions and authority. Ferguson said he would resist treating AI agents as autonomous actors with “wills and desires of their own,” a position that keeps attention on the people and organizations behind the automated system.

This distinction matters because an AI agent does not independently decide its purpose, access level, or operating environment. A company chooses whether an agent can read customer records, send messages, change settings, make purchases, or interact with other online services. Those design choices can shape the scope of potential consumer harm before the agent begins a task.

AI tools are increasingly being placed inside work and consumer services. For example, businesses are testing systems that can take actions across connected applications, while persistent workplace agents are designed to handle broader task flows. The most sensible approach is to treat each permission as a deliberate business decision, not as a default convenience setting.

Does the FTC chair’s statement create a new AI rule?

The FTC chair’s September 25 statement does not create a new AI-specific liability rule. Ferguson said the United States should use existing legal tools to address AI-related problems, and he suggested that FTC breach-disclosure authority could apply to AI developers in appropriate circumstances.

The significance is that federal scrutiny may not wait for Congress to enact a single comprehensive AI law before agencies examine alleged harms. A company that builds or deploys an agent may still face questions about the product’s conduct under laws and authorities that already exist, depending on the facts of a case.

That said, Ferguson’s remarks do not establish that every agent error creates liability, nor do they define a universal standard for developers. The legal outcome would depend on the specific conduct, the harm alleged, the instructions given to the system, and the authority the FTC chooses to use. Businesses should therefore avoid presenting automated decisions as beyond human oversight.

How do impersonation scams connect to AI agents?

AI agents can increase impersonation risk when they are used to automate persuasive messages, target users, or optimize the delivery of deceptive content. On September 24, 2026, the FTC announced an Advance Notice of Proposed Rulemaking that examines whether platforms’ ad-optimization tools can further impersonation scams and whether the agency should amend its Impersonation Rule, create a separate rule, or take non-regulatory action.

The FTC said consumers submitted more than 1 million imposter-scam reports in 2025 and reported nearly $3.5 billion in losses. The agency also said nearly 30% of people who reported losses said their first contact with a scammer occurred on social media, representing $2.1 billion in reported losses. The FTC’s announcement explains the scope of the agency’s inquiry.

These figures do not establish that AI caused every reported scam. They do show why the FTC is examining systems that can improve advertising delivery or automate contact at scale. Consumers should independently verify unexpected payment requests, account warnings, and customer-support messages, particularly when a message creates urgency or asks for credentials.

What should businesses do before deploying an AI agent?

Businesses should give AI agents the minimum access needed for a defined task and require human approval for high-impact actions. An agent that can draft a customer reply has a different risk profile from an agent that can change account details, purchase products, export customer data, or send external communications without review.

Businesses should use 4 basic controls before expanding an agent’s authority. They are written task boundaries, limited data access, approval checkpoints, and activity records. Those controls make it easier to identify who authorized an action, what information the agent used, and whether a human had a reasonable chance to stop an error.

1. Define the exact tasks the AI agent can perform and the actions it cannot perform.
2. Limit access to customer data, financial accounts, administrator settings, and external messaging tools.
3. Require human approval before an AI agent sends messages, spends money, changes records, or shares sensitive information.
4. Review activity records regularly and suspend the agent if behavior departs from approved tasks.

Businesses should stop and seek legal, security, or compliance advice before giving an AI agent access to regulated data, payment systems, identity records, or irreversible system controls. A short internal review cannot replace professional guidance when an automated workflow could create significant consumer, privacy, or financial harm.

What should consumers do before using an AI agent?

Consumers should treat an AI agent like a service that can act through connected accounts, not like a harmless search box. Before connecting an agent to email, cloud storage, shopping services, or calendars, review what information the service can access and what actions it can take without asking again.

Consumers should avoid giving AI agents passwords, financial account details, government identification numbers, medical information, or private material that does not need to be included in the task. Privacy settings can reduce exposure, but settings do not change the fact that an agent may need access to complete the work a user assigns.

AI agent access also creates a broader account-security concern. A reported case involving AI agent image exposure illustrates why users should carefully consider the data that automated systems can reach. The practical response is to use separate accounts where possible, revoke unused connections, and check an agent’s action history after completing important tasks.

What does the FTC’s scam rulemaking examine?

The FTC’s impersonation-scam rulemaking examines whether platforms’ ad-optimization tools can further scams that impersonate governments or businesses. The commission voted 2-0 to submit the Advance Notice of Proposed Rulemaking for Federal Register publication, and comments will be due 60 days after publication.

The FTC has not concluded that every ad-optimization tool enables fraud. The notice asks whether the existing Impersonation Rule should be amended, whether a separate rule is appropriate, or whether non-regulatory action would better address the issue. The FTC’s Advance Notice of Proposed Rulemaking sets out those questions.

The rulemaking matters for AI because automated targeting and optimization can affect how quickly deceptive content reaches likely targets. Users should remain cautious when a social media advertisement, browser alert, or direct message claims that an account is in danger. Fake support alerts can also lock a browser screen or pressure users into calling a fraudulent number, as seen in browser-based tech support scams.

What happens next for AI agent oversight?

AI agent oversight is likely to develop through agency actions, public rulemaking, product decisions, and individual enforcement matters rather than through Ferguson’s remarks alone. The FTC’s September 24 scam inquiry and Ferguson’s September 25 liability comments both show that the agency is examining how automated systems can affect consumer harm.

For developers, the central lesson is that labels such as “autonomous” or “agentic” are unlikely to end questions about product design and human direction. For businesses, the central lesson is to document why an agent received a permission, who approved it, and how the organization can detect and stop harmful behavior.

For consumers, the best protection remains limited access and careful verification. Do not allow an AI agent to act on a sensitive account simply because the product promises convenience. Disable unnecessary connections and contact the actual company through a trusted website or app if an agent, advertisement, or message asks for money, credentials, or urgent action.

FAQ

Can AI developers be liable for harm caused by an AI agent?

Yes, FTC Chair Andrew Ferguson said AI developers could be liable when agents they instruct cause harm. The statement reflects his view that responsibility should remain with the people and companies directing the technology rather than with the tool itself.

Did the FTC create a new AI agent liability rule?

No, the FTC chair’s remarks did not create a new AI-specific rule. Ferguson said existing legal tools can be used for AI-related problems, but future enforcement would depend on the facts and the authority the FTC applies.

Why are AI agents riskier than ordinary chatbots?

AI agents can be riskier because they may be connected to accounts, data, and services that let them take actions. A chatbot that gives an incorrect answer can be inconvenient, while an agent with broad permissions can create a larger privacy, financial, or security problem.

What permissions should an AI agent not receive?

AI agents should not receive unrestricted access to financial accounts, passwords, sensitive personal data, or irreversible system controls. Give an agent only the access required for a specific task and require approval before high-impact actions.

Is the FTC investigating AI-powered scams?

Yes, the FTC is seeking public comment on whether platforms’ ad-optimization tools can further impersonation scams. The agency has not concluded that all such tools enable fraud, but it is considering potential rule changes and other responses.