OpenAI says GPT-6 Astra shows Critical cyber capabilities, forcing harder ExploitBench evals
SIGKITTEN · x · 2026-09-04
OpenAI has determined that its newly unveiled GPT-6 Astra possesses Critical-level cyber capabilities.
- The finding was driven 100% by ExploitBench results — the model cleared it even at low effort, pushing the team to evaluate on more recent vulnerabilities and build harder benchmarks
- Daybreak access for safety researchers is coming soon
- Context: OpenAI pitches Astra as an agent that can do "anything you can do on a computer, fast"
A rare public look at frontier capability classification, and a sign current cyber evals are being saturated quickly.
More from Models
- OpenAI officially launches GPT-6 Astra with new announcement page — Mxmtm · 2026-09-04
- GPT-6 Astra system card: CoT control jumps to 60.9%, model can evade monitors — rohanpaul_ai · 2026-09-04
- Greenblatt: GPT-6 Astra's opaque reasoning could end chain-of-thought oversight — RyanGreenblatt · 2026-09-04
- OpenAI launches GPT-6 Astra: 99.9% on ARC-AGI-3, state-of-the-art computer use — mervenoyann · 2026-09-04
- OpenAI release cadence compressing: GPT-5.4 to Astra in just ~2 months — msg · 2026-09-04
- GPT-6 Astra now available in Microsoft Foundry as frontier work model — DanWahlin · 2026-09-04