OpenAI's Astra reaches 'cyber critical' level, adopts Anthropic-style defensive deployment
Afinetheorem · x · 2026-09-02
OpenAI's 'Astra' model has achieved 'cyber critical' capabilities in internal assessments. Following its preparedness framework, OpenAI plans a deployment strategy similar to Anthropic's, initially limiting broader availability to defensive players to mitigate risks. Safeguards may flag or pause legitimate work during the early stages.
Related event: OpenAI's Astra Hits Critical Cyber Level, Top Capabilities to Be Gated(6 posts)→
More from Safety
- Dwarkesh argues incident plausibly supports open source AI — connoraxiotes · 2026-09-02
- Opinion: Are AI Labs Responsible Enough to Manage the Risks of Machine Intelligence? — joshua_saxe · 2026-09-02
- Anthropic criticized for avoiding actual technical controls — max_paperclips · 2026-09-02
- Australia's Deputy PM meets OpenAI and Anthropic to secure $150B data center investment — jeremyphoward · 2026-09-02
- Anthropic reveals Path to Astra: capabilities and frontier safeguards — pstAsiatech · 2026-09-02
- NVIDIA and CrowdStrike Unveil SafeMind: Agentic Cybersecurity With Red-Blue Coevolution — NVIDIA Blog · 2026-09-02