Voluntary

UK AI Safety / AISI

UK AI Safety Institute / DSIT · London, UK · effective 2023 → ongoing

UK's principles-based, regulator-led approach plus the AI Safety Institute for frontier-model evaluation. Pro-innovation, sector-regulator enforcement rather than a single AI act (for now).

Your Council assistant — do it all here

You are interacting with an AI system.

▶ Run a compliance simulation Get compliant — 2-min check Get trained Get a signed System Card Full clause-by-clause crosswalk →

Who must comply

▸Frontier model developers (voluntary evals)
▸Firms under existing UK sector regulators

Penalties

Via existing sector regulators (e.g. ICO fines under UK GDPR).

Key obligations

Five principles

Safety, transparency, fairness, accountability, contestability.

Frontier evals

Pre-deployment safety testing with AISI for frontier models.

Sector rules

Comply via existing regulators (ICO, FCA, MHRA…).

Sectors in scope

Frontier AI labsFinanceHealthcarePublic sector

Threats & cybersecurity it addresses

Frontier model misuseSystemic capability riskSector-specific harm

Crosswalks — comply once, cover many

EU AI Act →NIST AI RMF →ISO 42001

CSOAI Layer 0 mapping

Charter Art. 2, 17, 21, 50

Governed MCP tools — open source, pip/npx install

⎇ csoai-governance-crosswalk-mcp