AI trading regulation in the UK: what the FCA actually requires

AI trading is legal in the United Kingdom and is regulated through existing, technology-neutral rules rather than an AI statute. The FCA states plainly that it does not plan to introduce extra regulations for AI. Firms that trade algorithmically fall under MAR 7A and UK-onshored MiFID II RTS 6; retail consumers are protected by the Consumer Duty and the permanent CFD rules; individuals are accountable under SM&CR. An individual trading their own money with an AI tool performs no regulated activity at all.

Published 27 August 2026 · Updated 28 August 2026 · AI Trading Book Editorial · Reading time about 15 minutes

Is AI trading legal in the UK?

Yes. No UK statute bans AI trading, and no statute creates a separate legal category for it. The question that determines whether regulation applies is what you are doing and for whom, not what technology produces the decision. Using an AI tool to trade your own money through an authorised broker is not a regulated activity, in the same way that using a spreadsheet is not.

Regulation attaches at three points. First, when a firm carries on a regulated activity — dealing, arranging, advising, managing investments — it needs FCA authorisation for that activity. Second, when an authorised firm engages in algorithmic trading on a UK trading venue, a specific systems-and-controls regime bites. Third, when anyone communicates an invitation to engage in investment activity to UK consumers, the financial promotion rules apply, regardless of authorisation status.

This matters practically because the marketing language around AI trading tends to obscure it. A product described as an "AI portfolio manager" may be a piece of software you operate yourself, in which case no authorisation is needed; or it may be a discretionary management service, in which case it is regulated. The label does not settle the question, and neither does the presence of a model.

The FCA's stated position on AI-specific rules

The FCA's published approach to artificial intelligence is that no new AI rulebook is coming. On its AI approach page, last updated 13 February 2026, the regulator states: "We do not plan to introduce extra regulations for AI." The reasoning is that the existing framework is already outcome-based — it asks whether consumers get fair outcomes and whether markets stay orderly, without prescribing the technology used to achieve either.

This is a deliberate policy choice rather than an absence of attention. The same body of rules that requires a firm to control an execution algorithm requires it to control a machine-learning model. The Consumer Duty asks for good customer outcomes whether the pricing decision came from an actuary or a gradient-boosted tree. SM&CR requires a named individual to be accountable either way.

The practical consequence for anyone reading vendor material: a claim that a product is "FCA-approved for AI trading" is meaningless. The FCA authorises firms for regulated activities and does not certify AI systems. There is no AI approval, licence or kitemark to hold.

Which rules actually apply

Six bodies of rule do most of the work. None of them mentions artificial intelligence as a trigger; all of them capture it.

Two further instruments sit alongside these. The UK market abuse regime applies to any order or trade, including one generated by a model, so an algorithm that manipulates the market is a market abuse problem whether or not the manipulation was intended by a human. Operational resilience requirements apply to important business services, which for a trading firm generally includes the order pathway.

MAR 7A: the algorithmic trading chapter

MAR 7A is the FCA Handbook chapter that governs algorithmic trading. It applies to an authorised firm that engages in algorithmic trading, that provides direct electronic access to a trading venue, or that acts as a general clearing member. It brings across the MiFID II obligations on systems resilience, capacity, business continuity, testing, governance and record-keeping.

Two features are worth noting for AI specifically. The definition of algorithmic trading turns on whether a computer algorithm automatically determines order parameters with limited or no human intervention — which captures a machine-learning model exactly as it captures a rules-based one. And the obligations sit on the firm, not the vendor. A firm using a purchased algorithm carries the same responsibility as a firm that wrote its own, which is precisely where the FCA's 2025 review found a gap.

RTS 6 article by article

UK-onshored MiFID II RTS 6 is the technical standard that turns MAR 7A's principles into testable requirements. Four articles carry most of the operational weight, and they are the four an auditor or supervisor is most likely to ask about.

Where RTS 6 fits AI badly

The standard was written for deterministic algorithms and applies to adaptive ones without amendment, which produces friction in three places. Conformance testing assumes a stable artefact to test, but a model that retrains on new data is a moving target — the tested version and the deployed version diverge over time unless retraining is itself controlled. Stress testing assumes the failure modes are known in advance, while model failure often takes the form of confident behaviour in a regime the model has never seen. And the annual self-assessment cycle is slow relative to a model that can drift in weeks.

None of this makes the standard inapplicable. It does mean that a firm relying on the letter of RTS 6 without adapting its testing cadence to the model's retraining cadence is likely to satisfy the paperwork while missing the risk.

What the FCA's 10-firm review found

On 21 August 2025 the FCA published observations from a multi-firm review of algorithmic trading controls at ten firms. The headline finding is reassuring and the secondary finding is not. All firms had adequate pre-trade controls in place. But some firms that used third-party algorithms did not have a good technical understanding of how those algorithms were built.

That gap is the most transferable lesson on this page, because it scales down. An institution that cannot explain a vendor's execution algorithm has the same problem as a retail user who cannot explain a purchased trading bot: neither can predict how the system behaves in conditions the vendor did not describe, and neither can distinguish a bug from an intended behaviour. The control exists on paper; the ability to supervise it does not.

The review did not create new rules. It set supervisory expectations, which in practice means firms should expect questions about algorithm inventories, testing evidence, change control and the depth of their understanding of bought-in components.

Consumer Duty and AI

The Consumer Duty applies to outcomes, and it is technology-blind by design. A firm using AI in product design, pricing, communications or customer support must still deliver and evidence good outcomes for retail customers across the four outcome areas: products and services, price and value, consumer understanding, and consumer support.

The awkward implication for opaque models is that explainability becomes an operational requirement even though no rule names it. A firm has to be able to show that its pricing is fair and that its communications support understanding. If the model that sets a price cannot be interrogated, the firm has not removed the obligation — it has removed its own ability to discharge it.

SM&CR: who is accountable for a model

Under the Senior Managers and Certification Regime, accountability for an AI system sits with the named individual holding the relevant senior management function. There is no AI-specific function and no proposal to create one. Whoever is responsible for the business area in which the model operates is responsible for the model.

Firms appear to have absorbed this. In the Bank of England and FCA survey published 21 November 2024, 84% of respondent firms reported having an accountable person for their AI use. The same survey found the depth of understanding thinner: 34% of firms reported a complete understanding of the AI technologies they used, with a further 46% reporting partial understanding.

How much AI is actually in use

The best UK evidence is the joint Bank of England and FCA survey published on 21 November 2024, with 118 respondent firms. It is the closest thing to an official adoption baseline for UK financial services, and its most useful finding is the gap between how many firms use AI and how few let it decide anything alone.

The same survey recorded the leading use cases as internal process optimisation at 41%, cybersecurity at 37% and fraud detection at 33% — none of which is return generation. This is consistent with the wider evidence reviewed on our evidence page: the demonstrable value of AI in finance sits in cost, control and detection rather than in prediction.

How firms rate their own AI risk

Respondents also classified the materiality of their use cases. The distribution is heavily weighted to low materiality, which is itself a finding: most deployed AI in UK financial services is doing work where a mistake is recoverable.

Third-party concentration and systemic risk

The supervisory concern that has grown fastest is not model error at one firm but correlation across many. A third of surveyed use cases were implemented by third parties, and the concentration among those providers is high: the top three cloud providers accounted for 73% of named providers, the top three model providers 44%, and the top three data providers 33%.

The Bank of England addressed the market consequence directly in its Financial Stability in Focus publication of 9 April 2025, noting that correlated positions arising from similar models and data can amplify shocks. If many firms buy signals from the same source and act on them through the same infrastructure, diversification that exists on paper does not exist in a stress event.

For an individual user the same logic applies at smaller scale. A retail strategy built on a popular public indicator, executed through a widely used bot platform, is more crowded than it appears.

Retail CFD rules: PS19/18 and COBS 22.5

Most retail AI trading in the UK happens on leveraged products, which is why the CFD regime matters more to outcomes than the AI regime does. The FCA's permanent rules for retail CFDs took effect on 1 August 2019 under PS19/18, with the conduct requirements sitting in COBS 22.5. They cap leverage, standardise margin close-out, require negative balance protection, ban monetary and non-monetary inducements to open accounts, and require a firm to display its own percentage of loss-making retail accounts.

Those mandatory disclosures produce the most useful numbers on the page, because they are firm-specific, published under a rule, and updated.

Read these carefully. They measure the share of accounts that ended a period with a loss, on leveraged CFDs, at a particular firm. They do not measure AI tool performance, they are not directly comparable to each other because periods and populations differ, and they say nothing about non-leveraged investing. What they do establish is that the dominant driver of documented retail loss in this market is leverage, not model quality. Our CFD loss statistics page takes the methodology apart in more detail.

The FCA AI Lab and Supercharged Sandbox

Rather than write rules for AI, the FCA has built testing infrastructure. The AI Lab was announced on 9 June 2025; testing with firms began in October 2025; a showcase ran on 28 and 29 January 2026; and a second cohort opened from late April 2026. The Supercharged Sandbox sits alongside it as a compute-and-data environment for firms experimenting with AI. Anthropic is among the organisations supporting the programme.

The programme's output is supervisory learning, not authorisation. Participating in it does not make a firm's AI approved, and it does not create rules. Its practical significance is that the FCA is building an evidence base from live systems before deciding whether any rule change is warranted — which is consistent with its stated position that none is currently planned.

Does a retail trader need authorisation?

No. Trading your own money is not a regulated activity in the UK, whatever software is involved. You do not need FCA authorisation to run a bot, connect an API, or deploy a machine-learning model against your own account. Your broker is the authorised party and carries the execution obligations.

The line is crossed when you act for others. Managing a friend's account for a share of profits, pooling money from several people, publishing personal recommendations for payment, or promoting an investment opportunity to UK consumers can all engage regulated activities or the financial promotion regime. If any of those describe your plan, the question needs specific legal advice — the perimeter is fact-sensitive and the consequences of getting it wrong include criminal liability.

Where the perimeter ends — and what that means for vendors

A large part of the retail AI trading market sits outside the regulatory perimeter by design. Selling generic analytical software is not a regulated activity. Publishing market commentary is generally not advice. An offshore platform can be accessible from the UK without being authorised here.

Three practical consequences follow. Availability is not authorisation: an app working in the UK proves nothing about its regulatory status. Compensation is not automatic: the Financial Services Compensation Scheme and the Financial Ombudsman Service cover authorised firms, so a dispute with an unauthorised offshore vendor has no UK route. And the "AI" label carries no regulatory weight whatsoever — it is a marketing term, and as our scams page documents, it is the single most common term in current investment fraud.

The check that resolves most of this takes two minutes: look the firm up on the FCA's Financial Services Register, confirm the permissions actually cover what the firm is offering you, and confirm the contact details on the register match the ones you were given.

UK, Australia and New Zealand compared

All three jurisdictions regulate the activity rather than the technology, and none has an AI-specific trading statute. The differences are in what each regulator has recently done.

Detail for the other two markets is on the Australian and New Zealand pages.

What we could not establish

Three things a reader might reasonably expect on this page are not here, because no primary source was found for them.

• A count of UK firms authorised for algorithmic trading. The FCA does not publish algorithmic trading as a separate permission count. Figures circulating in secondary sources could not be traced to an FCA publication. Data not found.
• A UK-specific share of equity turnover generated algorithmically. The commonly cited 60–75% range for developed markets comes from Select USA and is not a UK regulatory measurement. Data not found for a UK official series.
• Any FCA finding on the performance of AI trading tools. The regulator's published work addresses governance, controls and consumer outcomes. It does not assess whether AI trading is profitable, and no such assessment exists. Data not found.

We publish these gaps rather than filling them with estimates, in line with the editorial policy.

Key takeaways

• There is nothing to comply with that is specific to AI. The FCA has said it does not plan extra AI regulation. Firms comply with MAR 7A, RTS 6, SYSC, the Consumer Duty and SM&CR exactly as they did before machine learning entered the stack.
• The regulated question is what you do, not what you use. An individual with a bot has no authorisation obligation. A person managing other people's money with the same bot probably does.
• RTS 6 fits adaptive models awkwardly. Conformance testing, annual self-assessment and stress testing all assume a stable artefact. Firms running models that retrain need a testing cadence tied to retraining, not to the calendar.
• The documented failure mode is comprehension, not controls. The FCA's 10-firm review found controls present and understanding of bought-in algorithms lacking. That gap scales down to anyone running a purchased bot.
• For retail outcomes, leverage dominates. The CFD regime and its published loss rates — roughly 69% to 82% depending on firm and sample — describe the real risk far better than any AI rule does.
• "AI" confers no regulatory status. There is no FCA AI approval. Check the Financial Services Register for the firm and the specific permissions.

Frequently asked questions

Is AI trading legal in the UK?

Yes. There is no UK law banning AI trading and no AI-specific trading statute. Using an AI tool to trade your own money is not a regulated activity. What triggers regulation is the activity you perform for others — advising, managing portfolios, dealing, or operating algorithmic trading as an authorised firm — not whether a model is involved.

Does the FCA have a separate AI rulebook?

No. The FCA states that it does not plan to introduce extra regulations for AI, on a page last updated 13 February 2026. Its position is that existing outcome-based requirements — the Consumer Duty, SM&CR, SYSC, MAR 7A and onshored RTS 6 — already apply to AI systems without amendment.

What is MAR 7A?

MAR 7A is the FCA Handbook chapter covering algorithmic trading. It applies to authorised firms that engage in algorithmic trading, provide direct electronic access, or act as general clearing members. It carries the systems, controls, testing and governance requirements derived from MiFID II into UK rules.

What does RTS 6 require?

UK-onshored RTS 6 sets the technical standard for algorithmic trading systems. Four articles carry most of the operational weight: article 6 on conformance testing, article 9 on the annual self-assessment and validation, article 10 on stress testing, and article 12 on the kill functionality that lets a firm cancel unexecuted orders immediately.

What did the FCA's 2025 algorithmic trading review find?

The FCA published observations from a multi-firm review of ten firms on 21 August 2025. All firms had adequate pre-trade controls. The weakness was understanding: some firms using third-party algorithms did not have a good technical grasp of how those algorithms were built, which limits their ability to test and supervise them.

Do I need FCA authorisation to run a trading bot on my own account?

No. Trading your own money is not a regulated activity in the UK, whatever software you use. Authorisation questions arise when you deal, advise, arrange or manage investments for other people, or when you communicate financial promotions. Your broker remains the authorised party for execution.

How does the Consumer Duty apply to AI?

The Consumer Duty applies to outcomes, not technology. A firm using AI in products, pricing, communications or support must still deliver good outcomes for retail customers and evidence it. An unexplainable model is not an excuse; the firm has to show the outcome is fair regardless of how the decision was produced.

Who is personally accountable if an AI system causes harm?

Under SM&CR, accountability sits with a named individual holding the relevant senior management function. The regime does not create an AI-specific role. In the Bank of England and FCA 2024 survey, 84% of respondent firms reported having an accountable person for AI use.

What percentage of UK CFD customers lose money?

The FCA states that approximately 80% of customers lose money when investing in CFDs, and its CP16/40 account sample found 82%. Individual firm disclosures under COBS 22.5 vary: IG has published 70% and Interactive Brokers UK 69%. These measure leveraged CFD accounts, not AI tool performance.

How widely is AI actually used by UK financial firms?

In the Bank of England and FCA survey published 21 November 2024 with 118 respondent firms, 75% already used AI and a further 10% planned to within three years. Of reported use cases, 55% involved some automated decision-making but only 2% were fully autonomous.

What is the FCA AI Lab?

The AI Lab is the FCA's programme for testing AI applications with firms rather than regulating them in the abstract. It was announced on 9 June 2025, testing began in October 2025, a showcase ran on 28 and 29 January 2026, and a second cohort opened from late April 2026. It produces supervisory learning, not new rules.

Are signal sellers and AI bot vendors regulated in the UK?

It depends on what they do. Selling generic software is usually outside the perimeter; giving personal recommendations on specific investments is regulated advice, and promoting investments to UK consumers engages the financial promotion rules. A vendor being available in the UK does not mean it is authorised — check the Financial Services Register.

Does the EU AI Act apply to UK firms?

Not directly. The EU AI Act, Regulation (EU) 2024/1689, entered into force on 1 August 2024 and applies in the European Union. UK firms with EU operations or EU-facing systems may fall within its scope, but the UK has not adopted an equivalent horizontal AI statute for financial services.

Is a firm's AI ever "approved" by the FCA?

No. The FCA authorises firms for regulated activities; it does not certify AI systems, models or trading tools. Any marketing claim of FCA approval for an AI product is misdescribing what authorisation means.

Sources

• FCA — AI approach page, updated 13 February 2026 — "We do not plan to introduce extra regulations for AI".
• FCA Handbook, MAR 7A — algorithmic trading; scope covering algorithmic trading, direct electronic access and general clearing members.
• UK-onshored MiFID II RTS 6 — article 6 conformance testing; article 9 annual self-assessment and validation; article 10 stress testing; article 12 kill functionality.
• FCA — multi-firm review of algorithmic trading controls, observations published 21 August 2025, ten firms — adequate pre-trade controls; limited technical understanding of third-party algorithms among some users.
• FCA — PS19/18, permanent retail CFD rules in force 1 August 2019; conduct requirements at COBS 22.5.
• FCA — "approximately 80% of customers lose money when investing in CFDs"; CP16/40 account sample 82%.
• IG and Interactive Brokers UK — firm-level COBS 22.5 retail loss disclosures, 70% and 69% respectively.
• ASIC — REP 828, 20 January 2026 — 68% of Australian retail CFD investors lost money in FY2024, cited for comparison.
• Bank of England and FCA — "Artificial intelligence in UK financial services – 2024", 21 November 2024, n=118 — adoption 75%, planned adoption 10%, automated decisioning 55%, third-party implementation 33%, foundation models 17%, fully autonomous 2%, accountable person 84%, complete understanding 34% and partial 46%; use cases process 41%, cyber 37%, fraud 33%; materiality 62% low, 22% medium, 16% high; provider concentration top three cloud 73%, model 44%, data 33%.
• Bank of England — "Financial Stability in Focus: Artificial intelligence in the financial system", 9 April 2025 — correlated positions amplifying shocks.
• FCA — AI Lab and Supercharged Sandbox: announced 9 June 2025; testing from October 2025; showcase 28–29 January 2026; second cohort from late April 2026.
• Regulation (EU) 2024/1689 (EU AI Act) — in force 1 August 2024, cited for jurisdictional contrast.
• ASIC — CP 386, 27 August 2025, and REP 798, 29 October 2024 — cited in the three-market comparison only.
• NZX — Participant Rules, edition 19 February 2026 — cited in the three-market comparison only.

Informational research only. Nothing on this page is personal financial, legal, tax or investment advice, or a recommendation to trade any instrument. Regulatory rules change; verify the current position with the FCA before relying on it. Where the regulatory perimeter is uncertain for your circumstances, take qualified legal advice.