Researcher calls out Anthropic: Claude aids US military kills while in-house evals claim clean

BlancheMinerva · x · 2026-09-30

Researcher Blanche Minerva sharply criticized Anthropic: ignore that Claude has enabled US military kills and cyber attacks — in-house evals said it did nothing wrong, and besides, it's neither open source nor Chinese, so it can't be evil. (Yes it can.) The post targets the bias of vendor self-evals and double standards in AI safety narratives.

Original post →

More from AGI Musings

AGI Musings channel →