The rules that reach the action.
Seven instruments now bear on organisations that let AI systems act: two EU regulations, two UK conduct regimes, a PRA supervisory statement, a voluntary US framework and an international standard. This page maps each one from its primary source.
How to read it
A navigator, not legal advice. Four kinds of obligation, and the page would mislead if it set them all the same way.
Directly applicable legislation.
EU AI Act · DORA
A regulator's rulebook.
Consumer Duty · SMCR
Supervisory expectation or voluntary framework.
SS1/23 · NIST AI RMF
An international standard.
ISO/IEC 42001
Instrument by instrument
Seven instruments. What each one is, whom it covers, and what it asks of an action.
EU AI Act
Regulation (EU) 2024/1689 — the EU’s cross-sector rulebook for AI systems, in force since 1 August 2024 and re-timetabled by Regulation (EU) 2026/1744 of 8 July 2026. It covers providers and deployers of AI systems in the EU, with the heaviest obligations attaching to high-risk systems.
Sources: EUR-Lex, Regulation (EU) 2024/1689 (consolidated 27 July 2026) · EUR-Lex, Regulation (EU) 2026/1744 · accessed 1 September 2026.
Article 12(1) requires that high-risk systems technically allow the automatic recording of events — logs — over the system’s lifetime, and Article 12(2) requires that logging support the identification of risk situations, post-market monitoring under Article 72, and the deployer’s own monitoring under Article 26(5).
Article 26(2) requires deployers to assign human oversight to natural persons with the competence, training and authority to exercise it; Article 26(5) requires them to monitor operation and suspend use where risk appears; Article 26(6) requires automatically generated logs to be kept for at least six months.
Penalties run to EUR 35 million or 7% of worldwide annual turnover for prohibited practices (Article 99(3)), and to EUR 15 million or 3% for breaches of listed operator obligations, including Articles 16, 26 and 50 (Article 99(4)).
Regulation (EU) 2026/1744, in force 27 July 2026, deferred the high-risk application dates: standalone Annex III systems moved from 2 August 2026 to 2 December 2027, and product-embedded Annex I systems from 2 August 2027 to 2 August 2028.
| Provisions | Applies from |
|---|---|
| Prohibitions and AI literacy (Chapters I–II) | 2 February 2025 |
| General-purpose AI, governance and penalties provisions | 2 August 2025 |
| General application date (Article 113) | 2 August 2026 |
| Additional Article 5 prohibitions added by Regulation (EU) 2026/1744 | 2 December 2026 |
| High-risk: standalone Annex III systems (Article 6(2)) | 2 December 2027 — was 2 August 2026 |
| High-risk: product-embedded Annex I systems (Article 6(1)) | 2 August 2028 — was 2 August 2027 |
Treat older summaries with care. Many still cite the superseded 2026 and 2027 dates for high-risk obligations.
Decision records, refusal records and oversight evidence for the actions an AI system proposes.
- Every governed action is sealed in a signed, tamper-evident record.
- A refused action is recorded with the reason it was refused.
DORA
Regulation (EU) 2022/2554 on digital operational resilience for the financial sector. In force since January 2023; its obligations have applied since 17 January 2025 (Article 64).
It covers EU financial entities — credit institutions, payment and e-money institutions, investment firms, central counterparties, trading venues, central securities depositories, fund managers, insurers and reinsurers, among others (Article 2) — with a lighter regime for microenterprises.
Sources: EUR-Lex, Regulation (EU) 2022/2554 · accessed 1 September 2026.
DORA does not mention AI by name; AI tooling a financial entity relies on falls to be assessed under its provisions on ICT services. Article 28(1)(a) puts the principle plainly: a financial entity using ICT services remains fully responsible, at all times, for discharging its own obligations under the regulation. Responsibility does not move to the vendor.
Article 28(3) requires a register of information covering all contractual arrangements with ICT third-party providers, distinguishing those supporting critical or important functions, reported annually to the competent authority. Article 30(2)–(3) prescribes what those contracts must contain, including — for critical or important functions — unrestricted access, inspection and audit rights and mandatory exit transition periods.
Articles 17–23 impose a unified reporting regime for major ICT incidents, and Articles 31 onwards place critical third-party providers under direct EU oversight.
Records a firm can examine, and hand to its own auditors, without depending on the vendor being present or reachable.
- Records can be verified offline, without access to our systems.
FCA Consumer Duty
FCA Handbook rules made under the Financial Services and Markets Act and finalised in policy statement PS22/9 of 27 July 2022: a Consumer Principle — Principle 12 — requiring firms to act to deliver good outcomes for retail consumers, with the detailed rules in PRIN 2A and non-Handbook guidance in FG22/5. It is not a statute. It covers FCA-regulated firms serving retail markets.
Sources: FCA, PS22/9 — A new Consumer Duty (27 July 2022) · accessed 1 September 2026.
The Duty is built around four outcomes — products and services, price and value, consumer understanding, and consumer support — with cross-cutting rules that apply across the whole relationship between the firm and the people it serves.
The rules do not distinguish an outcome produced by staff from one produced by software. The Duty applied to new and open products and services from 31 July 2023 and to closed books from 31 July 2024; both dates have passed, so it now applies across open and closed books alike.
Decision records for the moments where an AI-driven interaction touches a retail outcome.
- Every action the platform governs is checked before it executes; an action that fails its checks does not execute.
Senior Managers and Certification Regime
The UK’s individual-accountability regime: statutory in origin, with the detail in FCA and PRA rules. It has three components — the Senior Managers Regime, the Certification Regime and the Conduct Rules — and it applied to the banking sector from March 2016, insurers from December 2018, all solo-regulated firms from December 2019 and benchmark administrators from December 2020.
Sources: FCA, Senior Managers and Certification Regime · FCA, CP25/21 and PS26/6 (SM&CR review) · accessed 1 September 2026.
The regime attaches accountability to named individuals. Six Individual Conduct Rules apply to staff, including rule 2 — due skill, care and diligence — and rule 6, added with the Consumer Duty and in effect since July 2023, which requires individuals to act to deliver good outcomes for retail consumers; senior managers carry four further conduct rules on top.
Accountability sits with the named senior manager for their area of responsibility, however the work inside it is done. An automated system does not move it.
The regime is being reformed: PS26/6 (21 April 2026) finalised phase-one changes that took effect in stages between 24 April and 1 September 2026, and a second phase depends on Treasury legislation, with further consultation expected later in 2026.
Oversight evidence: a record of which actions were held for a person, and of what that person approved.
- Agents run under quotas and spend caps, with human approval gates on the actions that matter.
PRA SS1/23 — model risk management
A Bank of England supervisory statement setting out the PRA’s expectations for model risk management in banks: first published 17 May 2023, effective from 17 May 2024, with the current version published 23 April 2026 after low-impact amendments. A supervisory statement states expectations; it is not a Rulebook rule or a statute.
Sources: Bank of England, SS1/23 — Model risk management principles for banks (current version 23 April 2026) · accessed 1 September 2026.
It applies to UK-incorporated banks, building societies and PRA-designated investment firms that hold internal model approval for regulatory capital — not to every UK bank.
It sets five principles: model identification and risk classification; governance; model development, implementation and use; independent model validation; and model risk mitigants. The principles apply to all models used to inform business decisions, in-house or vendor-supplied, regardless of technology; the sub-principles expressly include identifying and managing the risks of AI and machine learning in modelling, and allocate responsibility for the framework to the appropriate senior management function holder.
Independent validation depends on reproducibility, and a governance layer is itself something a model-risk function will want to test.
- Same input, same verdict, every time.
- The kernel applies deterministic mathematical governance; if its own checks cannot be discharged at start-up, it does not serve.
NIST AI Risk Management Framework
The US National Institute of Standards and Technology’s AI RMF 1.0 (NIST AI 100-1), released 26 January 2023. It is voluntary, non-sector-specific and use-case agnostic: any organisation may adopt it, and none must. A companion Generative AI Profile (NIST AI 600-1) followed on 26 July 2024.
Sources: NIST, AI Risk Management Framework hub · NIST AI 100-1 (full text) · accessed 1 September 2026.
The framework’s core is four functions — govern, map, measure and manage — each broken into categories and subcategories, with govern designed as a cross-cutting function that informs the other three.
For an organisation using the framework to structure its AI risk work, govern is where the questions about authority over actions, and the records they leave, get answered.
A concrete governing behaviour to point at when the framework asks how failure is handled.
- When the governance layer cannot decide, the action is refused, not waved through.
ISO/IEC 42001
ISO/IEC 42001:2023, Information technology — Artificial intelligence — Management system, published December 2023 as Edition 1 by ISO/IEC JTC 1/SC 42. It specifies requirements for establishing, implementing, maintaining and continually improving an AI management system, and ISO describes it as the first AI management-system standard. It is written for organisations of any size that develop, provide or use AI-based products or services, across industries.
Sources: ISO, ISO/IEC 42001:2023 catalogue record · accessed 1 September 2026.
The full text is purchased from ISO. This page cites the public catalogue record only, and gives no clause numbers for that reason.
The records a management system runs on.
- The evidence a period produces can be handed to a third party and verified with Admissible absent.
The thread
The common demand is a record. Each regime asks, in its own way, for records of what an AI system did and for humans answerable for it.
The record is an input to that question, not a conclusion about it.
What a record contains
The proposal, the checks it faced by name, the verdict, the moment, and the seal.
Check one yourself
The shipped verifier passes on the true record and fails on the altered copy, offline.
Why the action, not the answer
An answer can be corrected. An action must be answered for, and these instruments ask who answered.
Start here
Bring the obligation you answer to
We will show you the actions it reaches, the verdict each one gets, and the record it leaves behind.