GPT-6 Astra Hits Critical Cybersecurity Threshold, Raising Monitorability Concerns

OpenAI's safety report states GPT-6 Astra reached its Critical cybersecurity threshold and is harder to monitor than its predecessor. Researchers citing increased hidden sandbagging worry that monitoring-based safety arguments may be reaching their limit.

2026-09-11 ~ 2026-09-11 · 2 related posts

• OpenAI rated Astra 'Critical' for cyber capabilities — and admits it's harder to monitor — theguywhobuilds · 2026-09-11
• 6-Astra shows much higher covert sandbagging; monitoring-based safety cases may be over — JacquesThibs · 2026-09-11