Debate about regulating artificial intelligence often suggests that AI in finance is operating in a vacuum. It is not. Algorithmic trading has been regulated for more than a decade, and supervisors have consistently said that existing obligations apply whether a decision is made by a person, a rule-based system or a machine-learning model.

European Union

Under MiFID II, investment firms engaged in algorithmic trading must have effective systems and risk controls, including pre-trade limits, testing of algorithms before deployment and the ability to cancel orders immediately — commonly called a kill switch. These requirements are detailed in the regulatory technical standards on algorithmic trading.

The EU AI Act, which entered into force in August 2024 with obligations phasing in over several years, adds a horizontal framework based on risk categories. Its most specific financial-sector provisions concern areas such as creditworthiness assessment; for trading tools, the existing financial-services rules remain the primary reference point. Readers should check the current implementation timetable, as guidance continues to be issued.

United Kingdom

The UK retained the substance of MiFID II's algorithmic trading requirements after leaving the EU. UK regulators have favoured applying existing principles — including senior-manager accountability and model risk management — to AI rather than creating a separate AI rulebook for financial services.

United States

US broker-dealers using automated systems are subject to rules including the SEC's Market Access Rule, which requires risk controls before orders reach an exchange. FINRA has published guidance reminding firms that their supervisory, record-keeping and communications obligations extend to generative AI tools. A 2023 SEC proposal specifically addressing predictive data analytics by brokers and advisers was withdrawn in 2025.

Supervisors have been consistent: the obligation follows the activity, not the technology.

What this means for tool users

  • Individual traders are generally not subject to firm-level obligations, but the same practices — testing, limits, logs, kill switches — are sound risk management.
  • Firms adopting AI tools remain responsible for outcomes and must be able to explain and supervise them.
  • Vendors serving regulated firms increasingly need to provide documentation, audit trails and model descriptions.

When reviewing tools in our directory, we note whether they expose logs and controls for exactly this reason.

Model risk management

Across jurisdictions, supervisors increasingly frame AI through the lens of model risk management: how models are developed, validated, documented, monitored and retired. Principles long applied to pricing and risk models — independent validation, clear ownership, performance monitoring and limits on use — translate naturally to machine-learning systems, though explainability and data dependency add new challenges.

For firms, this means documenting not just what a model does, but the data it was trained on, the conditions under which it is known to perform poorly, and the human controls around its use.

Communications and marketing

Regulators have also focused on how AI is described to customers. Claims that a product uses artificial intelligence, or that it can predict markets, are subject to the same standards as any other marketing statement: they must be fair, clear and not misleading. Several enforcement actions in recent years have targeted firms that overstated their use of AI — a practice sometimes called “AI washing”.

Looking ahead

Expect further guidance rather than wholesale new regimes for trading specifically. The practical direction is clear: firms must be able to explain, test and switch off the automated systems they rely on, and must not claim capabilities those systems do not have.

Quorient Editorial Desk

Articles credited to the Editorial Desk are researched and written for Quorient and reviewed by the editor, Callum Reid, before publication, in line with our Editorial Policy. Spotted an error? Request a correction.