Governance
Safety frameworks edition by edition, who controls the labs, and 5 external commitments and laws.
- Document history
Frontier safety frameworks
22 editions of 7 safety frameworks from Anthropic, Google DeepMind, OpenAI, xAI, Meta and Microsoft. Each edition shows its dates, what replaced it, quoted provisions and the changes recorded since the previous edition. The page can be downloaded as a research starter, a file of the claims, sources and open questions.
Read the document histories → - Control and commitments
Who controls the labs, and what they committed to
Dated timelines of who holds which control right over which organisation, through which instrument, and whether it was announced or in effect. The commitments page covers 5 external commitments and laws, with signatory or refusal records for 25 organisations. It quotes their provisions, gives their dates and lists the published frameworks that link themselves to them. A commitment or legal duty is never evidence of compliance.
- OpenAI’s control structure, 2015–2026 62 records
- Anthropic’s control structure, 2021–2026 31 records
- Research question
Frameworks, commitments and legal obligations, 2023–2026
From voluntary frameworks to legal obligations: how did frontier-AI developers’ self-governance change, 2023–2026? A chronology and 202 existing records, each linked to its original source, with a list of what the records cannot answer. No essay or conclusion is added.
Open the research question →