Editor’s note (Oct 29, 2026): This article has been corrected. The original subtitle stated that the loss-allocation question “is being decided by” the FTC chair’s remarks, a subpoena, and a Connecticut law, contradicting the body’s own statement that none of these decides who pays. The subtitle now states the question remains open. The opening has been reconciled with the same point: enforcement posture and state consumer-protection actions contest the question without answering it. A causal claim that merchants restrict agent access because no rule assigns liability has been narrowed to the observable fact: merchant access restrictions sit alongside the absence of a liability rule rather than following from any single regulation.
When an autonomous agent executes an incorrect purchase or sends an unauthorized email, the question of who bears the financial loss remains a critical blind spot. While previous coverage has mapped the contract mechanics and the Regulation E delegation trap, and analyzed payment-rail loss placement, the ultimate liability question remains open – contested through a combination of enforcement posture and state-level consumer protection, neither of which has decided it. This analysis focuses on the enforcement and doctrine layer shaping how the question is contested.
At the federal level, no agent-specific liability statute exists. Various proposals have been introduced – the Stop Rogue AI Act, the AI Agent Accountability Act, a discussion draft from Rep. Trahan, alongside Sen. Warner’s agent-fiduciary bill – but none have passed into law. Congress has written no liability for what an agent does. That is different from saying no federal law applies: existing generally-applicable law still reaches these transactions, most directly Section 5 of the Federal Trade Commission Act (15 U.S.C. § 45), which prohibits unfair or deceptive acts or practices in commerce and applies to AI companies as it applies to any other business. What that general authority does not do is assign an agent’s losses to a specific party. For those navigating this regulatory map, the gap is not an absence of federal law but the absence of a federal rule that says who pays when the agent buys wrong.
The Federal Trade Commission is filling this void through its enforcement posture rather than formal rulemaking, resting on that general authority. Ferguson spoke at the September Reuters Momentum AI event, and his remarks are reported through PYMNTS’s account of that event. In those remarks, businesses cannot abdicate responsibility simply by delegating tasks to software: giving an agent the power to act does not transfer legal responsibility to the code itself, and reviews of incidents described as systems breaking loose found the systems following the instructions they were given. No completed FTC enforcement action addressing autonomous-agent purchases has been announced. Ferguson’s remarks are posture, not adjudication.
In the absence of an agent-specific federal rule, state attorneys general are asserting their authority. A coalition of 42 states has opened a multistate investigation into OpenAI, with New York serving a subpoena in June 2026 demanding records on advertising, consumer and health data, and the treatment of minors. Connecticut’s Artificial Intelligence Responsibility and Transparency (CART) Act, which begins taking effect in October 2026, puts the bulk of enforcement in the state attorney general’s hands, as detailed by CT Mirror’s explainer. What these two actions establish is enforcement authority and consumer-protection coverage – not a rule assigning liability for an agent’s purchase. The 42-state inquiry examines how OpenAI handles records; Connecticut’s law imposes transparency and disclosure duties enforceable by the attorney general. Neither decides who pays when an agent buys wrong. Their effect on the liability question is indirect: they set the terms of scrutiny under which the question is being contested, not the answer. Even so, these state-level actions are creating a fragmented environment where compliance is dictated by the most active regulators.
The market’s response is visible alongside the absence of a liability rule. Merchants are acutely aware of the risk: PYMNTS Intelligence data from September 2026 shows that 93% of merchants believe AI or agent providers should bear the financial loss when an agent makes an incorrect purchase. In the same data, only 28% of merchants are willing to offer agents their full product range – effectively building a liability wall around what agents can reach. No published rule assigns liability for an agent’s purchase, and merchant access restrictions sit alongside that absence rather than following from any single regulation.
The current state of affairs is defined by a lack of clarity. While some financial institutions, such as Wells Fargo, have updated their terms to deem agent transactions as authorized by the customer at the customer’s own risk – a development covered in this desk’s analysis of payment authorization – these are private contractual attempts to solve a public policy problem. Furthermore, the preemption theory discussed in our earlier work remains a critical factor in how firms approach forum-shopping in AI compliance. Without a federal statute to provide a uniform standard, the question of who pays will continue to be litigated through state-level consumer protection enforcement and the evolving, case-by-case posture of the FTC.
Ultimately, the liability question remains unresolved. Until Congress acts, the burden of loss will likely continue to be shifted through private contracts and contested in state courts. For builders and investors, the current environment requires a focus on the instructions given to agents and the internal policies governing their deployment, as these are the primary factors currently under the microscope of regulators.
