Anthropic: a downloadable Chinese AI model can now build working hacks on its own

ross2000 · reddit · 2026-09-30

Anthropic warns that a Chinese AI model anyone can download can now autonomously produce working hacks/exploits on its own. The company flags this as concerning evidence that openly available model weights are lowering the barrier to autonomous cyber-offense, raising questions about open-source model misuse.

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

Original post →

More from Safety

Safety channel →