The AI itself owes you nothing, because it has no legal personality. Any duty belongs to the person or firm offering the advice, and how strong that duty is depends heavily on whether they are registered, where, and what they promised you. As of November 2026, how these duties apply to advice generated by autonomous AI is one of the central unresolved questions in this field.
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.
Who owes the duty when the advice comes from an AI?
The humans and businesses behind it. An AI system cannot be sued, so the law has to find a natural person or a company to hold responsible (the Law Commission of England and Wales' July 2025 discussion paper on AI and the law). We cover that in more depth in Can an AI agent be sued, or own property?
Using AI does not let a professional off the hook. The Law Commission notes that lawyers who file AI-generated material without checking it remain responsible for their own statements, however the content was produced. Our reading: the same logic should apply to a firm that passes on AI-generated investment recommendations under its own name.
Better-run frameworks build this in. A reference framework published under Singapore's central bank says humans should remain accountable for consequential decisions, with human review for escalated actions (the SAFR reference paper). In its case studies, institutions assign each agent a named, accountable human owner. It is an industry reference approach, not regulatory guidance. But in our view it shows where serious institutions are heading: every agent has an owner.
Why does registration matter so much?
Because registration is what ties advice to a regulator, a rulebook and a complaints route, and it is one of the first things US regulators tell investors to check. In a January 2024 alert on AI and investment fraud, the SEC's investor-education office, together with state securities regulators and the brokerage industry's self-regulatory body, warned about platforms, often unregistered, promoting AI trading systems with unrealistic promises (the SEC investor alert on AI and investment fraud).
Their advice was simple. Confirm you are dealing with a registered investment professional and a registered platform. Check the free public database for registration and disciplinary history. Treat a lack of registration as a reason to dig further before investing.
Automated advice is not new to regulators. A 2026 academic paper notes that robo-advice, software that allocates portfolios automatically, has operated subject to suitability rules (an academic paper on agent-to-agent finance). The open question is what happens when the software stops following a fixed model and starts deciding for itself.
| Who is behind the "AI adviser" | Is there likely to be a duty? | What to check |
|---|---|---|
| A registered firm that uses AI in its advice | Yes: the firm's duties do not vanish because it used AI | The firm's registration and what it says the AI does |
| An unregistered app selling an "AI trading system" | Unclear, and regulators warn about exactly this | Registration, promises of returns, who you would complain to |
| A general-purpose AI chatbot you ask about investing | Our reading: weak, unless it is offered as personal advice | Its terms; whether it claims to advise you |
| An agent you set up yourself | You are the one deciding; the duty runs mostly to you | Your own limits and logs |
The whole table is our reading, not settled law, and the third and fourth rows are the least settled. They are where the legal fight ahead is likely to happen.
What does “duty of care” actually mean for AI advice?
In negligence law, a duty of care is a legal duty to take reasonable care to avoid foreseeable harm to someone. With AI, the hard part is deciding who in the chain owes it, and what "reasonable" means.
The Law Commission paper sets out three problems. First, it may be unclear which parties in an AI supply chain owe a duty to the person eventually harmed. In the paper's example, a healthcare provider using an AI tool is likely to owe its patients one; the duties of upstream developers are far less clear. Second, even a small risk of harm can be foreseeable, but a defendant who took all reasonable precautions may still escape liability.
Third, automation bias, the tendency of people to over-trust a machine's output, can carry an AI's error into a human's decision. The paper asks a question that applies directly to advisers: at what point should a professional decline to follow an AI system's recommendation?
Product law does not fill the gap for money. The EU's new Product Liability Directive treats software, including AI systems, as a product and applies to products placed on the market after 9 December 2026. But it covers death, injury, property damage and some data loss, and it excludes pure economic loss (the directive's adopted text). A bad AI investment recommendation that costs you money is not, on its own, covered.
What could go wrong?
The worst case is not a bad recommendation. It is a bad recommendation with nobody legally answerable for it.
The liability gap. The Law Commission gives an example that fits this topic exactly. A company's autonomous AI sends false statements to investors to persuade them into agreements, and no employee reviews them first. Proving the company knew, or was reckless, could be very hard. If neither can be shown, there may be no criminal liability.
Unregistered "AI advisers". Regulators specifically warn about unlicensed firms selling AI trading systems with too-good-to-be-true returns. An unregistered seller may owe you very little in practice, whatever its marketing says.
Bad data behind confident advice. The SEC alert warns that AI-generated information can be inaccurate, outdated or manipulated, and should not be the only basis for an investment decision.
Over-trust on both sides. Automation bias affects advisers too. A human who rubber-stamps an AI's output may add a name to the advice without adding any real judgement.
Liability scattered across the stack. The 2026 academic paper notes that responsibility may be fragmented across users, deployers, model providers and wallet operators, and lists contractual allocation among the controls. Until contracts do that, a claim can bounce between parties.
The trust takeaway
Ask who stands behind the advice, not how clever the AI is. A duty of care is only as real as the registered, solvent, reachable person who owes it. Before acting on an "AI adviser", check:
- Is the firm behind it registered, and does it appear on the regulator's public register?
- Does it say in writing that it is giving you personal advice, or only "information"?
- Does a qualified human review the AI's recommendations before they reach you?
- Which regulator oversees it, and what is the complaints route?
- Can it show you why it recommended what it did?
For projects rather than individual investors, the same principle applies to strategy. Decisions about token design, treasury policy and launch planning should sit with people who can be held to account for them. That is the work of Alcyone, TrustSwap's strategy and advisory practice. In our view, AI can help with the modelling, but accountable people should make the calls.
Agent trades can also create tax obligations that no adviser takes on for you: see Do I owe taxes on trades my AI agent made?