FCA Asks AI Firms to Warn Users About the Limits of Chatbot Financial Advice

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Britain's Financial Conduct Authority has asked the companies behind general-purpose AI models to warn users about the limits of the financial information and advice those models generate. Jessica Rusu, the FCA's chief data, information and intelligence officer, made the request at Bloomberg's UK Policy Series, and the text of the speech, published by the regulator, asks for "clear, proportionate and consistent disclaimers." Rusu said those warnings may work best at the point where AI-generated financial information reaches the user. No deadline was attached. No new rule was written.

That gap between a request and a rulebook is the exact space where the AI trading bot sub-niche now operates, and it is why we spent part of our 2026 review cycle re-reading the disclosure pages of every automated system on our roster. We benchmarked live automation against the Ellington AI trading platform and a spread of algorithmic platforms, expert advisors and copy trading systems, and the disclosure problem the FCA is describing is not theoretical for us. It is the first finding in almost every vendor audit we run.

Rusu tied the request to two figures. Four in five less-experienced investors aged 18 to 40 have used AI for help with investing, she said, but only 40% of consumers understand they have no recourse if they act on general-purpose AI advice and it goes wrong. Put those two numbers side by side and you have the retail risk profile of 2026 in a single line: near-universal adoption sitting on a 60% misunderstanding rate.

Then there is the perimeter question, which is where this gets uncomfortable for anyone with money in an automated strategy. Last week, FCA chief executive Nikhil Rathi told lawmakers that a chatbot offering general information, without a personal recommendation that leads to a transaction, is not wholly inside the regulated perimeter. Finance Magnates reported at the time that the regulator plans to automate its own supervision with AI while holding off on rules for chatbots (Finance Magnates).

Read that carefully. General information sits outside. A personal recommendation that leads to a transaction sits inside. An AI trading bot is, by design, a machine that converts general information into transactions, and nobody has yet said which side of the line that lands on.

What did the FCA actually ask for?

Strip the language back and the request is narrow. The FCA wants model providers to put warnings in front of users. It has not written a rule, set a deadline, or built a licensing regime for chatbots. Rusu also set out the regulator's response to the Mills Review, published in July, on how AI could reshape retail financial services by 2030, and said the FCA is embedding six of its seven recommendations into live work (Finance Magnates). Instead of building a new public tool for financial capability, as the review proposed, the FCA will build on the first TechSprint of the International Organization of Securities Commissions (IOSCO), run with its AI Lab, to help investors spot scams and question unreliable AI answers.

The scam pressure is not hypothetical. Australia's securities regulator took down more than 19,400 scams in its latest fiscal year as AI deepfakes targeted investors (Finance Magnates). One regulator, one year, one country.

Table 1: What the FCA asked for versus what is actually in force

Sources: FCA speech text as reported by Finance Magnates, 2026. Firm-level permissions should be checked on the FCA Register and, for Australian entities, on ASIC's registers.

Why does this matter for AI trading bots?

Because the brokers did not wait for the framework. At least 10 retail brokers and platform vendors connected AI agents to live client accounts in the first half of 2026, and Anthropic's Claude was named in nine of those launches, according to an FM Intelligence study (Finance Magnates). Our team logged those 10 launches against our own vendor checklist, and the split in how much authority each agent gets is the single most important thing a retail trader can examine before handing over an account.

Interactive Brokers lets Claude draft trades that clients must approve before an order goes out (Finance Magnates). At the other end of the spectrum, eToro lets investors delegate trades to AI agents that run inside separate portfolios (Finance Magnates). Those are not the same product and they are not the same risk. One is an assistant that waits for a human. One is a delegated mandate that does not.

Table 2: How much authority does the agent actually get?

Free Download: FCA-Aligned AI Chatbot Advice Checklist: 12 Checks Before You Trade on Bot Signals
A due-diligence checklist that helps you verify whether an AI trading bot's chatbot advice is generic or personalised, how it discloses its limits, and whether its regulatory status, disclaimers, and execution flow can be trusted with real capital.
Get the FCA Advice Checklist

Sources: Interactive Brokers via Finance Magnates, eToro via Finance Magnates, FM Intelligence study. Fields marked "varies" are not disclosed consistently across the 10 launches, so verify with each provider directly.

Not sure which AI trading bot fits your strategy? Try Ellington: The AI Trading Platform for 2026

This link is an affiliate partnership - see our editorial policy for details.

What does an AI trading bot actually do with your money?

Strategy specification is the first thing we check, and it is where most vendors are weakest. In plain English, a properly specified bot should publish its instrument universe, its signal, its entry and exit logic, its position sizing rule, and the conditions under which it stands down. When a vendor will not publish the stand-down conditions, we treat that as a finding rather than a marketing gap, because the stand-down rule is the part of the specification that protects your account when the market regime changes.

Then there is the backtest versus live-trade gap, which is always there and always real. Backtests run on clean data with a fixed rule set. Live markets deliver slippage, gaps and API latency. We do not publish a single "average" gap, because it is meaningless across strategy classes; the honest answer is that the size of the gap depends on the strategy, the instrument and the broker's execution quality, and the only way to size it for your own account is to ask the provider for their live-to-backtest disclosure and then verify it against your own fills. If a vendor will not hand that over, treat the headline backtest as marketing.

Broker compatibility and API integration is the next filter. An agent that can only talk to one broker is a single point of failure, and the FCA's own good and poor practice guidance lists dependence on a single model or vendor as an area for improvement. That guidance creates no new requirements, the regulator said, but it does point to a named senior manager for each AI use case, and it includes a case study of a firm that had to redesign a product after the model behind it was retired.

That case study should worry every bot subscriber. If your bot wraps a single large language model and that model is retired, updated or repriced, the bot's behaviour can change without anyone touching the strategy file. The FCA calls it vendor concentration. We call it a strategy you never agreed to.

Do the disclaimers actually reach anyone?

Here is the part of the FCA's request that we think the market is reading too optimistically. Rusu said warnings may work best at the point where AI-generated financial information reaches the user. For a chatbot, that point is a screen, and a human is looking at it. For an AI trading bot, that point is an API call at 3am, and no human is looking at anything. A disclaimer rendered at the point of delivery is a human-facing control bolted onto a machine-facing workflow. It will satisfy a compliance file. It will not slow down a single order.

The near-miss point is the more interesting one, and it is the piece of the speech we would put at the top of the page. Rusu said the duty to report major incidents to the FCA "must extend to significant AI failures and near-misses," and she described near-miss reporting for agentic AI as both a regulatory and a national security priority. For a discretionary trader, a near-miss is a trade you almost took. For an automated account, a near-miss can be a stop that almost failed to trigger, a hedge that almost did not fill, or a model that almost sized a position far larger than the risk budget allowed. The retail account holder has no visibility into any of it, and there is currently no obligation for the bot vendor to tell them.

How much authority should an agent have over your account?

The FCA says it will work with industry, other regulators and frontier labs on a framework covering agent identity, delegated authority, accountability and execution. It gave no timetable. Rusu's framing was blunt: "Agentic systems therefore require continuous, conditional assurance." A system that is safe at launch may behave differently at scale or when the underlying model changes.

That is the right frame for a retail trader evaluating an AI trading bot. We reduce agent risk to four questions. Who is the agent? What is it allowed to do without asking? Who is accountable when it does something else? And how does it execute? If a vendor cannot answer all four in writing, the bot is not ready for a funded account, regardless of what the backtest says.

Table 3: The adoption numbers behind the FCA's request

How does Ellington compare?

On the concrete dimension of vendor concentration, the structural difference is multi-strategy automation. Ellington runs more than one strategy under a single portfolio-level risk layer, which spreads execution and model dependency across a wider surface than a single-strategy bot that wraps one underlying model. That is not a marketing claim, it is an architecture claim, and it is the direct answer to the risk the FCA's own guidance now flags as an area for improvement.

Compare that with the two live-agent models we logged in the first half of 2026. Interactive Brokers keeps a human in the loop on every drafted trade, which is safe but slow. eToro delegates execution inside segregated portfolios, which is fast but concentrated. Ellington's multi-strategy automation sits in a different lane because the portfolio risk control is applied across strategies rather than inside one, which is where a single-model bot has no answer at all.

What should retail traders do next?

Three practical steps. First, ask the vendor for their live-to-backtest disclosure and their model dependency statement in writing, and walk away if either is missing. Second, check the provider's permissions on the primary register, whether that is the FCA Register or ASIC's registers, before you fund anything. Third, test the disengagement path first, not last. If you cannot stop the bot cleanly and confirm that no orders remain live, you do not control the account.

The FCA's request is a warning shot, not a rule. But the direction of travel is clear, and the bots that will survive the next 18 months are the ones that can already answer the four questions above in writing.

Not sure which AI trading bot fits your strategy? Try Ellington: The AI Trading Platform for 2026

This link is an affiliate partnership - see our editorial policy for details.

Try Ellington: The AI Trading Platform for 2026

Try Ellington: The AI Trading Platform for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

Does the FCA's request change how AI trading bots are regulated?

No. The FCA framed the disclaimers as a call to AI providers and set no deadline, and the accompanying guidance creates no new requirements. The regulator has said a chatbot offering general information, without a personal recommendation that leads to a transaction, is not wholly inside the regulated perimeter.

Can I run an AI trading bot on a UK account if the provider is not FCA-authorised?

The perimeter question is unresolved for general information, so the safest approach is to verify the provider's permissions directly on the FCA Register before funding. If the vendor cannot point you to a register entry, treat that as a red flag.

What happens if the API connection drops mid-trade?

That depends entirely on the vendor's fail-safe design, and it is one of the four questions we insist on before any funded test. Ask for the documented behaviour on connection loss, and confirm who the named senior manager is for that AI use case, since the FCA's guidance now points to one per use case.

Do any of the live agent launches let the AI move my money?

No. None of the launches in the FM Intelligence study let an agent deposit, withdraw or move client money. Authority is limited to drafting, approving or executing trades inside a client account or a segregated portfolio.

Which model do most broker AI agents use?

Anthropic's Claude was named in nine of the 10 broker and platform launches that connected AI agents to live client accounts in the first half of 2026. That level of concentration is exactly what the FCA's guidance flags as an area for improvement.

Is a disclaimer at the point of delivery enough for an automated bot?

No, and this is the gap we think the market is missing. A disclaimer works when a human reads it on a screen. In an automated bot, the point of delivery is an API call, often overnight, and no human is present to read anything.

How do I check whether a bot depends on a single model or vendor?

Ask for the model dependency statement in writing and compare it against the FCA's good and poor practice guidance, which lists single-model or single-vendor dependence as an area for improvement. A vendor that cannot name its fallback model has no fallback.

What counts as a near-miss in an AI trading bot context?

A stop that almost failed to trigger, a hedge that almost did not fill, or a position that was almost sized far beyond the risk budget. The FCA wants near-miss reporting for agentic AI, but today there is no obligation for a bot vendor to disclose any of it to the account holder.

Does Ellington fit the FCA's framework better than a single-strategy bot?

On the vendor concentration dimension, yes. Ellington's multi-strategy automation applies portfolio-level risk control across more than one strategy, which is a structural answer to single-model dependence. A single-strategy bot wrapping one model cannot make the same claim.

Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

More in this category: Trading Industry News.