OpenAI says its Astra cybersecurity model hit Critical threshold under Preparedness Framework
burny_tech · x · 2026-09-03
OpenAI officially previewed Astra, an upcoming model for cybersecurity that it says has reached the Critical threshold under its internal Preparedness Framework — its highest capability tier. The company outlined how it evaluated the model, how safeguards have advanced alongside capabilities, and what it plans to keep improving. The repost wraps this in a snarky meme about how AI companies talk about safety, but the substance is the announcement itself.
Related event: OpenAI Previews Astra as gpt-6-astra Traces Multiply Ahead of Launch(85 posts)→
More from Fun
- Tiny model frontier mapped: GLiNER 2.5 leads <350M, Kev family tops two brackets — multimodalart · 2026-09-22
- A volunteer park ranger at Crater Lake just got cold-emailed by an AI — AaronBergman18 · 2026-09-22
- Chinese teacher and students build 4-meter water rocket that launches, separates and returns — TansuYegen · 2026-09-22
- China mass-produces transparent LED film for glass, but UV aging may leave ghost patches — TansuYegen · 2026-09-22
- Local MLX vs hosted API: two AI agents battle in Chrome's T-Rex game — usamawahabkhan · 2026-09-22
- Who grades the graders? The recursive problem haunting agent evals — Kyrannio · 2026-09-22