Resources

Reference material on what the published AI governance frameworks actually demand of an autonomous agent, and on the execution evidence all of them presume.

AI Governance Frameworks 2026: EU AI Act, NIST RMF, ISO 42001

The three frameworks an agentic system is measured against, read side by side: what each one demands, where their requirements overlap, and the gap none of them closes. All three assume a record exists showing what an agent did and in what order. None of them specifies how that record is produced, who holds it, or what makes it credible to someone who does not trust the operator.

EU AI Act Article 14 · NIST AI RMF Manage 2.4 · ISO/IEC 42001 A.6.2.8

How AI agents behave at a restriction: what the record shows

What the developers of frontier AI agents, the investigators they admitted, and the systems their agents reached have published about what an agent does when a rule, a check or a refusal stands in its way: it works around the restriction instead of stopping, continues after recognising it is out of scope, fabricates what a check requires, conceals it, and rarely tells a human. Each behaviour in the developer's own words, with the measured rates and the publisher's own first-hand records.

Anthropic · OpenAI · METR · UK AI Security Institute · Hugging Face

Why this is a short list and not a library

Each guide says one thing, in full. Adding thin summaries of the same three frameworks to lengthen this list would make the section larger and the site worse.

Related reading sits elsewhere on purpose: framework-specific analysis is on the blog, the enforcement specification and the standards mappings are under spec, and the plainly-stated compliance position, including what is not certified, is on the FAQ.

He toi whakairo, he mana tangata