The big AI labs' safety push could come with a competitive advantage

The article reports on recent safety-related announcements from large AI labs, and on calls by their executives to slow model development. It quotes PitchBook senior analyst Harrison Rolfes, who said such announcements can look performative and described the activity as AI safety 'theater.' The excerpt concentrates on that assessment and on one proposed arrangement.

The article describes a possible model in which the big labs would agree to have their frontier models tested by independent evaluation firms. It names METR, Redwood Research and Apollo Research as examples of such firms. It also cites an essay by Anthropic CEO Dario Amodei proposing independent safety evaluators embedded inside companies. Both are presented as proposals or possibilities rather than as arrangements already in place.

The headline frames the safety push in competitive terms, stating that it could come with a competitive advantage for the labs involved. The article sets Rolfes's characterization of the announcements as theater alongside the evaluator proposals involving METR, Redwood Research and Apollo Research. It attributes the 'theater' assessment to Rolfes and quotes no other analysts in the excerpt. No legislation or government action is mentioned in the excerpt, and the excerpt does not report evaluation results, a timeline, or any lab's commitment to the arrangement described. The excerpt also does not state whether the evaluation firms named have been formally engaged.