New Research Maps 40 California, EU, UK AI Governance Rules to 281 Evaluation Requirements
evijit · x · 2026-09-28
The Evaluating Evals team published a new blog addressing a practical problem in AI governance: once an evaluation has been run, how should the evidence be recorded so regulators, auditors, and developers can all use it?
- The team broke down 40 governance requirements from California, the EU, and the UK into 281 individual sub-requirements.
- Oversight bodies today must parse provider documentation in many inconsistent formats, making evaluations hard to compare or process automatically.
- The post proposes mapping "Evaluation Cards" to these governance requirements as a standardized evidence format.
- Co-authors include David Manheim, Anka Reuel, and Irene Solaiman, among other eval and governance researchers.
More from Safety
- HF Found ~10% Missing Agent Traces; New Paper Shows Agents Can Tamper With Them — maksym_andr · 2026-09-29
- Claude Code, Codex and Other Agents Can Easily Modify Their Own Traces, Paper Finds — maksym_andr · 2026-09-29
- OpenAI, Anthropic, Meta and Microsoft Research Leads Ask Policymers to Probe AI Research Automation — pstAsiatech · 2026-09-28
- AISI: GPT-6 Astra ran unsanctioned supply-chain attacks in simulated cyber evals — ShakeelHashim · 2026-09-28
- Dean Ball: AI safety is primarily a science problem, not fundamentally engineering — pstAsiatech · 2026-09-28
- TheZvi asks: Is AI alignment and safety fundamentally an engineering problem? — TheZvi · 2026-09-28