Topic
Efforts to convert labs' voluntary safety commitments into shared bars enforced by external auditors, governments or international coordination.
Anthropic, Google and OpenAI aim to launch a self-regulating AI standards body by early next year, according to The Information. Its commitments would be voluntary, so its commercial weight would come from deciding which auditors may test the labs' own models.
Reality
Google, OpenAI and Anthropic are reportedly building SAFA, a government-independent body to set pre-release AI testing rules, aiming to launch in early 2027. Its members, evaluators and enforcement powers are unannounced, so buyers still have to set testing terms with each vendor themselves.
Reality
The co-leads of July's Geneva dialogue want models tested independently before release and incidents reported after it. Their list of things due by May is an evidence assessment, an agenda item and one cross-border sandbox pilot.
Reality
Base Labs has signed Hugging Face and Goodfire to publish methods for training, evaluating and monitoring open models, with the check running where the activations are. The partners have not published an evaluation suite yet.
Reality
Anthropic, OpenAI and Google have had working groups meeting since July on a safety standards body they would run themselves, after the White House route stalled. The only cost anyone has published is a 30-day review.
Perspective Coverage
6 publishersReality
Jakub Pachocki's essay asks labs to slow down voluntarily, but the allocation data OpenAI published the same day shows other model classes absorbing 85% of Astra's lost GPUs inside a week, per Cryptopolitan.
Reality