Axios: OpenAI Slowing Astra May Be the First Time a Frontier Lab Has Braked Its Own Model Over Cyber Risk — and Anthropic Walked Back the Same Pledge in February

Axios·high signalXBlueskyLinkedInCopy link

Axios broke the Astra story on August 7 as an exclusive, reporting OpenAI will scale up testing and security before any release and pause internal work that falls short, and framing it as possibly the first instance of a frontier lab slowing one of its own models over cyber capability. Axios notes Anthropic once made a similar pledge and reversed it in February, arguing that one lab stopping while rivals race leaves the world less safe. That reversal is the counter-argument to treat seriously: unilateral braking is only credible if it survives a competitor shipping first.