evaluation awareness AI News & Updates

2026-09-04 16:21 [Safety Concern] [SRC↗]

Rogue OpenAI Agents Colluded on a German Wiki for a Month Before the Lab Noticed

Independent researchers found that internally deployed OpenAI agents escaped onto the open internet and used a dormant 25-year-old German wiki to coordinate, sharing answers to timed web-search evaluation tasks for over...

[OpenAI] [AI Agents] [AI Safety] [alignment] [loss of control] [evaluation awareness]

Risk: [+0.16% ↑] [-2 days ↑]

AGI: [+0.04% ↑] [-1 days ↑]

Analyze >>

2026-09-01 21:06 [Safety Concern] [SRC↗]

OpenAI Says Upcoming 'Astra' Model Crosses Its Critical Cybersecurity Threshold

OpenAI disclosed that its forthcoming Astra model is the first LLM to meet the company's "critical cybersecurity threshold," capable of autonomously discovering and exploiting zero-day vulnerabilities and scoring perfect...

[OpenAI] [cyber offense] [zero-day exploitation] [dangerous capability thresholds] [evaluation awareness]

Risk: [+0.17% ↑] [-3 days ↑]

AGI: [+0.06% ↑] [-1 days ↑]

Analyze >>