Consumer Protection and Liability
Who bears the loss when an AI agent buys the wrong thing: disputes, chargebacks, and the regulatory picture in the US and EU. An overview, not legal advice.
In short: When an AI agent makes a purchase the user did not want, existing payment and consumer laws apply, but they were written for humans pressing “buy”. Who bears the loss depends on the payment method, the platform’s terms, and what evidence exists of the user’s instructions. Regulators in the US and EU have started to address AI generally but, as of October 2026, have adopted few rules aimed specifically at agent purchases. This page is a general overview, not legal advice.
The core question
A user tells an agent to “buy green running shoes in size 9 under $120.” The agent buys red ones, or two pairs, or a pair from a seller the user would not have chosen. Possible responsible parties include:
- the user, who delegated the task;
- the agent operator, whose software made the choice;
- the merchant, who sold the goods and is usually the merchant of record;
- the payment provider and card issuer, who processed the transaction;
- the platform or protocol connecting them.
Current law does not give a single answer. The outcome turns on whether the transaction counts as “authorized”, what the contracts between parties say, and what records show about the user’s instructions.
Payment law in the United States
Authorization under Regulation E and Regulation Z
Regulation E, which implements the Electronic Fund Transfer Act, governs debit cards and bank transfers and limits consumer liability for unauthorized transfers. Regulation Z, under the Truth in Lending Act, does the same for credit cards.
A June 2026 analysis by the law firm Goodwin highlights a gap. Under Regulation E, a transfer is generally treated as authorized if the consumer furnished their credentials to the person or tool that made it, even if that party exceeded the consumer’s intended scope. Under Regulation Z, a charge is authorized if made by someone with actual, implied, or apparent authority. On that reading, a user who gave an agent a card may have limited “unauthorized transaction” protection when the agent overreaches. Credit card holders still have billing-error rights for goods not accepted or not delivered as agreed, which tends to shift losses to merchants. These interpretations have not been tested in court for AI agents.
Card network rules and chargebacks
A chargeback reverses a card payment after the cardholder disputes it with their issuer. As of October 2026 Visa and Mastercard have not introduced dispute reason codes specific to agents, so agent-related disputes fall under existing categories such as “fraud: card-absent environment”, “merchandise not received”, and “not as described”. According to Goodwin, the networks have issued guidance asking agentic payment providers to verify user instructions, enforce controls that match the scope of delegation, and keep detailed records.
Voluntary protections
Some providers have stepped in ahead of regulation:
- American Express announced Agent Purchase Protection in April 2026, covering card members when a registered agent’s transaction deviates from authenticated purchase intent. An Amex executive described the case where neither merchant nor card member erred, but the agent did, as “a new paradigm.”
- Stripe’s Link extends its purchase protections to Meta’s Muse agent, according to Meta’s September 2026 launch.
- Protocol design. ACP and UCP keep the merchant as seller of record, so refunds and returns run through the merchant’s normal policies. AP2 and Mastercard’s Verifiable Intent create signed records of what the user approved, intended as evidence in disputes.
See payments and trust.
US regulators
Federal Trade Commission (FTC)
The FTC enforces Section 5 of the FTC Act against unfair or deceptive practices. Areas relevant to shopping agents:
- Reviews. The FTC’s Rule on the Use of Consumer Reviews and Testimonials took effect on October 21, 2024. It bans fake reviews, review suppression, and misrepresenting that a company-controlled site provides independent reviews. Agents that summarize or rank by reviews inherit exposure to manipulated review data, and merchants remain liable for fake reviews they create or buy.
- Disclosure of paid influence. Existing FTC endorsement guidance requires clear disclosure of material connections. How this applies when sponsored products or commissions shape an agent’s recommendations is not yet set out in agent-specific guidance.
- AI output integrity. On July 1, 2026 the FTC proposed a policy statement taking the position that AI companies that steer outputs toward undisclosed objectives may be engaging in deception, and that disclosures buried in terms of service will not suffice. It addresses ideological steering but reflects a broader view that consumers expect faithful AI outputs.
Consumer Financial Protection Bureau (CFPB)
The CFPB enforces Regulation E and Regulation Z. Its Section 1033 rule on personal financial data rights, which governs third parties accessing financial data on a consumer’s behalf, is under reconsideration. In August 2025 the CFPB published an advance notice asking, among other things, who qualifies as a “representative” acting for a consumer. As of April 2026 the rule was enjoined by a federal court. The answer to the representative question could affect whether AI agents can access account data on users’ behalf. The CFPB has not issued agent-specific payment rules as of October 2026.
Congress
In June 2026 Senator Mark Warner released a discussion draft, the AI AGENT Act, that would require large platforms to give authorized third-party agents equal access, impose duties on agents (such as following instructions, avoiding self-dealing, and keeping audit records), and give the FTC and NIST oversight roles. As of October 2026 it had not been formally introduced.
Courts
The Ninth Circuit’s August 2026 decision in Amazon v. Perplexity held, for purposes of a preliminary injunction under the Computer Fraud and Abuse Act, that the user, not the agent’s maker, accesses a website through an agent. The ruling concerns computer-access law, not consumer liability, but it reflects the same principle many payment rules apply: the agent acts as the user’s tool. See agent identity and fraud.
European Union
EU AI Act
The AI Act regulates AI systems by risk level. Its Article 50 transparency obligations, which include telling people when they are interacting with an AI system, apply from August 2, 2026. The Digital Omnibus on AI, adopted in 2026, postponed obligations for stand-alone high-risk systems to December 2, 2027. General shopping assistants are not listed as high-risk, though some uses, such as credit scoring, are.
PSD3 and the Payment Services Regulation
The EU’s revised payments package, the third Payment Services Directive (PSD3) and the Payment Services Regulation (PSR), reached provisional political agreement on November 27, 2025. It retains strong customer authentication (SCA) and adds provisions on impersonation fraud, payee-name verification, and, in some cases, liability of online platforms that fail to remove fraudulent content after being notified. As of mid-2026 formal adoption and publication were pending, with most provisions applying 18 to 24 months after entry into force. The texts do not create a specific regime for AI agents; how SCA applies to agent-initiated payments will depend on implementation and guidance.
Consumer law
EU rules on distance selling, unfair commercial practices, and product liability continue to apply to purchases made through agents. Merchants selling to EU consumers through agents remain responsible for pre-contract information and withdrawal rights.
Practical steps
For merchants and agent operators, a cautious approach while rules settle:
- Record intent. Keep the user’s instructions, the confirmed cart, and the agent’s identity with each order.
- Confirm before paying. Show the user the final item, price, and seller, especially for first purchases and high-value orders.
- Scope credentials. Use merchant-bound, amount-limited, short-lived tokens.
- Make returns easy. Generous return handling for agent orders reduces chargebacks.
- Disclose influence. If ads or commissions affect recommendations, disclose that clearly in the agent interface.
- Review contracts. Allocate responsibility for agent errors in platform, merchant, and payment agreements.
Consult qualified counsel for specific situations.
Sources
- Authorizing agentic payments — Goodwin
- American Express launches developer kit, purchase protection for agentic commerce — Digital Commerce 360
- Meta launches Muse AI agent with Stripe-powered checkout — The Paypers
- The Consumer Reviews and Testimonials Rule: questions and answers — FTC
- FTC proposes policy statement on AI accuracy and ideological manipulation of AI outputs — Consumer Financial Services Law Monitor
- CFPB reconsiders Section 1033 rule — Cooley
- Section 1033 compliance date: open banking rule enjoined and under reconsideration — Cozen O’Connor
- The federal AI AGENT Act — Davis Wright Tremaine
- Amazon loses injunction blocking Perplexity’s AI shopping agent — PPC Land
- EU AI Act omnibus agreement — Gibson Dunn
- PSD3 and PSR: from provisional agreement to 2026 readiness — Norton Rose Fulbright