Cryptocurrency exchange Coinbase has launched an AI-powered investment adviser capable of producing personalised financial recommendations inside its app, in a development that could intensify scrutiny of how existing financial regulation applies to genAI tools.

Most genAI systems are careful to state that they are not expert advisers of any kind and that their responses should not be treated as professional investment advice. And for good reason – between hallucinations, bias and poor contextual understanding of any given user’s actual situation, having a chatbot make actual financial decisions for you is a scary thought.

With that in mind, Coinbase Advisor is described by the company as an AI-powered, SEC-registered adviser that can help users assess market news, explore investment ideas and build portfolio strategies through a chat-style interface.

According to Coinbase, the tool is designed to turn user goals and questions into personalised recommendations using Coinbase data, market information, user inputs, current holdings and risk preferences. The company says users can ask questions such as how market news affects them or request help building a portfolio, with the tool then producing recommendations that can be reviewed before any action is taken.

Beyond this, the crypto firm says the service can support strategies across products including crypto assets, equities, derivatives, automated dollar-cost averaging and USDC lending, with prediction markets expected to be added in future releases.

The company is attempting to position the product as a way to make investment guidance more widely available, arguing that the kind of support traditionally associated with financial advisers can now be built directly into its app – again though, distinctions have to be made. Financial advisers curate support tailored to each person’s financial position and beyond this, they don’t hallucinate – sure, they might give you bad advice, but at least they can be then compelled to give you a reason or rationale.

How is this legal?

The launch marks a big moment for the use of AI in financial services because Coinbase Advisor is not being presented simply as a chatbot providing general information. The product is offered through Coinbase Advisors, LLC, a separate Delaware-registered company that is registered with the US Securities and Exchange Commission as an investment adviser and with the US Commodity Futures Trading Commission as a commodity trading adviser.

In practice, this means the legal status sits with Coinbase Advisors, LLC, rather than the AI model itself. The AI system operates as the customer-facing interface for a regulated advisory service, while the firm behind it takes on the obligations attached to that regulatory status.

Coinbase’s Form CRS says Coinbase Advisors provides automated investment advisory services through a digital interface. It also states that the firm provides non-discretionary investment advice, does not manage client accounts, and limits its advice to digital assets, equities and derivatives.

The distinction between discretionary and non-discretionary advice is central to how the product works. Coinbase Advisor can analyse information and generate recommendations, but it cannot make trades on a user’s behalf without approval. Coinbase says every action requires explicit user permission, with customers retaining final control to review, edit and authorise transactions.

In simple terms, the model can suggest what a customer might do, but the customer still has to choose whether to act. Once the customer confirms an action, Coinbase Advisor can transmit the resulting order or order information to the relevant trading service provider.

That structure is key to the product’s legal footing. Rather than waiting for a new AI-specific law, Coinbase appears to have built the service inside existing US financial regulation by registering a corporate entity as an investment adviser and commodity trading adviser, then using AI as the delivery mechanism for the advice.

For investment advice, the relevant US framework is built around the registration and duties of the adviser. Registered investment advisers are subject to obligations including acting in the client’s best interest and managing conflicts of interest. Commodity trading adviser registration applies where a firm provides advice for compensation or profit on products such as futures contracts, options on futures, retail off-exchange forex contracts or swaps.

This does not mean the AI itself is licensed in the way an individual adviser might be. Instead, it means a regulated firm is using an AI system to deliver an automated advisory service.

To put it simply, the firm is the adviser, the chatbot works for the firm and that’s legally okay, apparently.

The risks

Coinbase’s own disclosures set out a number of risks attached to the service, many of which point to the core challenge of applying existing financial regulation to genAI.

The company warns that Coinbase Advisor may produce outputs that are wrong, incomplete, inaccurate, biased, false, inconsistent, unsuitable for a user’s purposes, or out of date. Its risk disclosures also state that the tool may misinterpret, oversimplify or hallucinate user questions or inputs.

Financial advice is a high-stakes use case so this really is a point that can’t be over-laboured. A hallucination in a general chatbot may produce misleading information – a hallucination in an investment adviser could influence a user to make a real financial decision. It’s also probably a fair leap of logic to say that someone who uses a free AI tool as opposed to paying a financial adviser (aka, the majority of us) is probably more likely to not be in a buoyant enough fiancial position to bear the brunt of a really bad financial decision.

So hallucinations are one risk. The second is arguably just a thorny. Coinbase says the tool may lack context on a user’s wider circumstances, including their overall financial position, risk tolerance or goals. The service primarily uses information available to Coinbase, such as balances, transaction history, information provided through the interface and suitability questionnaire responses.

That means Coinbase Advisor may not consider a user’s bank accounts, pensions, debts, income, expenses, other investments or financial obligations outside Coinbase. The company’s disclosures state that any recommendation may be unsuitable when viewed in the context of the user’s full financial situation.

This is one of the most important limitations. A user may receive a personalised recommendation based on what Coinbase can see, but that does not necessarily mean the advice is suitable when judged against their wider financial life.

Saving the best for last, there are some some pretty wild potential conflicts of interest at play here. Coinbase says Advisor may suggest, highlight or facilitate access to products and services offered by Coinbase entities. While Coinbase Advisors says it does not receive transaction-based compensation from trading activity connected to the service, other Coinbase entities may earn fees, spreads, interest, rewards or other compensation when users act on recommendations.

Recommended reading

• Report: 81% of Brits Have Never Heard of Stablecoins
• Deel To Launch Stablecoin Wages for UK Workers
• Crypto Platform Drift Drained of $280M in DPRK-Linked Raid

This comes directly from Coinbase’s own Advisor Risk Disclosures, under Section 5.1, “Affiliated products and revenue”.

The disclosure goes on to say:

“While Coinbase Advisors does not receive transaction-based compensation in connection with your trading activity, when you use these products:

The product also raises questions around accountability. Coinbase states that users remain solely responsible for reviewing and verifying transaction details before confirming a transaction. This includes the asset, side, size, price, order type and fees.

That may be legally important, but it also creates a difficult consumer protection issue. If an AI system shapes the user’s understanding of the market, constructs the strategy and presents the suggested action, the fact that the user clicks to confirm does not remove the influence of the system on the decision.

The risk is not simply that AI could make a mistake. It is that a conversational system may create a sense of confidence, fluency and personalisation that makes users more likely to trust an output, even where the underlying recommendation is incomplete, poorly contextualised or unsuitable.

The next phase of AI regulation in financial services may not simply be about whether a tool is legal, but whether existing rules are strong enough for the way consumers actually experience and rely on AI-generated recommendations.