Anthropic red team warns a downloadable Chinese AI model can now build working hacks on its own

ross2000 · reddit · 2026-09-30

Anthropic's Frontier Red Team warns that a Chinese AI model anyone can download can now autonomously build working hacks. The Reddit post flags it as worrying news, though it links no further details — the core claim is an open-weights model crossing a cyber-offense capability threshold per Anthropic's own red team assessment.

Related event: Anthropic Warns Open-Source Chinese Model Can Autonomously Build Hacking Tools(2 posts)→

Original post →

More from Safety

Safety channel →