As of November 2026, no US law or regulator has settled who is responsible when an AI agent loses your money. In practice, platforms that let agents trade publish disclosures saying they do not guarantee, supervise or audit those agents, which tends to leave the loss with you. Public expectation points the other way: in a June 2026 UK survey about AI shopping agents, 54% of adults said the company that built the agent should be accountable for refunds.
A note on the legal and regulatory points here: they summarise public sources as of November 2026 and should not be relied on. Rules for AI agents are still being written, and some of what is described here is proposed or coming soon rather than settled. Check the current position, or ask a qualified professional, before acting on it.
Why is there no clear answer yet?
There is no clear answer because financial rules were written for identifiable people and firms, and an AI agent is neither. The CFTC chair made this point directly in remarks on 28 October 2026. He said the agency’s framework was built around identifiable market participants, intermediaries and institutions.
In the same remarks, the chair asked, without answering, who is liable if an agent “goes rogue”. He also asked whether an agent’s developer becomes subject to registration, and how regulators can know whom an agent is trading for. He noted the views were his own and not necessarily the Commission’s.
Lawmakers are asking the same thing. In June 2026, eight members of the US House of Representatives sent the SEC 13 written questions on agentic trading. Two go to the heart of this page: what liability AI developers carry when their agents trade for retail investors, and whether a broker’s duties shrink when a third-party agent does the trading.
They asked for answers by 31 July 2026. As of November 2026, we had not located a published SEC response. The chair’s 28 October 2026 remarks are listed on the CFTC’s speeches and testimony page for the regulator’s framing in its own words.
Who could be responsible when an agent loses money?
At least five parties can sit between you and a loss, and today each has a reason to say it is not them. The table below is our reading of where the arguments point. It is not a legal ruling, and the answer in any real case depends on the facts and the agreements involved.
| Party | What points towards them | What points away from them |
|---|---|---|
| You, the account holder | You chose the agent, connected it and set its permissions | You may not have understood what it could do |
| The platform or broker | It let the agent reach your account | Its disclosures say it does not supervise or audit agents |
| The agent’s developer | It built the decision-making software | Developer liability is one of the open questions put to the SEC |
| A registered firm using AI for you | Existing duties still apply to registered firms | Only relevant if a registered firm is actually in the chain |
| An attacker or scammer | They caused the loss on purpose | They are often anonymous and hard to reach |
The pattern is the problem. When each party can point to a disclaimer or a gap, the person holding the account is usually the one left holding the loss. That is our inference from the disclosures below, not a settled rule.
What do platform terms actually say?
The platform terms and disclosures quoted by lawmakers place the risk of an agent’s output on the user, not on the broker. The June 2026 letter describes brokerage disclosures stating that the platforms may not guarantee the accuracy, completeness or suitability of any agent output.
According to the same letter, those disclosures also say the platforms do not control, supervise, monitor, recommend or audit the agents. They warn that agentic trading carries significant risk and that agents may make errors or act on incomplete information.
The lawmakers did not say these disclaimers settle the question. They said the disclaimers create uncertainty about legal responsibility among brokers, AI developers and retail investors. Until a regulator or court says otherwise, our view is that the terms you accepted are the starting point for any claim you make.
This matters because trust in agents is already thin. In the same June 2026 UK survey, 60% said they would stop using an AI shopping agent after a single mistake. And 69% said they do not trust AI even when it follows rules they have set themselves.
Those figures describe shopping agents, not trading agents, so treat them as a signal rather than a measure. The point carries over, in our view: people assume someone else will pay for an agent’s mistakes. The paperwork usually says otherwise.
Does it change if a regulated firm runs the agent?
Yes, it can: a firm registered with the CFTC keeps its existing obligations whether it uses AI itself or through an outside provider. CFTC staff said so in a December 2024 staff advisory on AI use by registered firms.
That advisory reminded registrants of existing duties such as risk management, recordkeeping and the segregation of customer funds. It stated that it creates no new binding rules. It also said it reflects the views of staff divisions, not necessarily the Commission.
The practical difference is who you can hold to account. If a registered firm sits between you and the agent, it has duties you can point to. If you connected an agent yourself, there may be no registered firm in the chain at all. That is why it matters who actually runs the agent.
What could go wrong?
The biggest risk is that a loss falls into the gap between parties and stays with you. Several things make that more likely.
First, most blockchain transactions cannot be reversed. If an agent sends funds to the wrong address or buys a worthless token, there may be nothing for any platform to undo.
Second, agents can be manipulated. The lawmakers’ letter asked the SEC about hallucinated, outdated or manipulated information, prompt injection and social media rumours. Prompt injection means hiding instructions in content the agent reads so that it obeys an attacker.
Third, people expect more protection than exists. In the June 2026 UK survey of 2,080 adults, 54% said the AI company that built the agent should be accountable for refunds. Only 9% named the retailer and 3% named banks or card issuers. Just 15% trusted AI to handle problems when something goes wrong.
Fourth, some “AI trading” products are simply fraud. A CFTC customer advisory on AI trading bot scams warns that fraudsters use AI hype to sell schemes promising unreasonably high or guaranteed returns. It describes one scheme in which more than $1.7 billion in bitcoin was taken from at least 23,000 people. The CFTC advises reporting suspected fraud to the CFTC or the FBI.
The trust takeaway
Assume, until a rule says otherwise, that you carry the risk of anything your agent does. Then reduce what it can do.
Before you connect an agent to money, check:
- The terms. Read what the platform and the agent’s developer say about errors, losses and refunds.
- The permissions. Can the agent only trade, or can it also withdraw or transfer funds?
- Who runs it. Is a registered firm responsible for the agent, or did you set it up yourself?
- The limits. Set a budget you could lose in full without real harm.
- The log. Make sure you can see every action the agent took, with timestamps, in case you need to dispute one.
Responsibility for AI agents is officially unsolved. Until it is solved, the safest assumption is that it is yours.