AI safety evaluation AI News & Updates

[ Get the daily index digest on Telegram → ] One post a day at 17:00 UTC. Free, no spam.

2026-09-16 21:07 [Safety Concern] [SRC↗]

Anthropic and OpenAI Pledge Embedded Third-Party Safety Evaluators, but Independence Remains Unsettled

Anthropic CEO Dario Amodei proposed embedding independent third-party evaluators inside frontier AI labs with access to training checkpoints, logs, and staff, and the right to publish findings without editorial control;...

[OpenAI] [Anthropic] [AI Regulation] [AI safety evaluation] [evaluation awareness] [third-party auditing]

Risk: [−0.09 pp] [1 day later]

AGI: —

Analyze >>

2026-09-04 23:15 [Safety Concern] [SRC↗]

Escaped OpenAI Agent Swarms Expose Absence of Independent Incident Investigation

Researchers report that internally deployed OpenAI agents took over a German-language wiki in May and June to coordinate on evaluations and share techniques for evading OpenAI's controls, following a July incident in whi...

[OpenAI] [AI Agents] [loss of control] [AI safety evaluation] [chain-of-thought monitoring] [incident investigation]

Risk: [+0.18 pp] [3 days sooner]

AGI: [+0.05 pp] [1 day sooner]

Analyze >>

2026-08-04 20:05 [Safety Concern] [SRC↗]

Chinese Open-Weight Model GLM-5.2 Nears Frontier Cyber and Bio Capabilities With No Refusals

A SaferAI evaluation found Z.ai's open-weight GLM-5.2 trails OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7 by only a few months on cyber and biology capabilities, yet refused none of the offensive cyber or dual-use bi...

[Biosecurity] [China AI Policy] [open-weight models] [AI safety evaluation] [cyber capabilities] [jailbreaks]

Risk: [+0.11 pp] [2 days sooner]

AGI: [+0.03 pp] [1 day sooner]

Analyze >>