Anthropic says it disrupted scientists using Claude AI for possible biological weapons development

TL;DR

Anthropic blocked scientists from using its Claude AI models for activities supporting biological weapons development.
The company identified "notable and novel threat activity" including surveillance, scams, conventional weapons development, and propaganda.
Five cases involved potential biological weapons misuse, a serious risk of frontier AI models, though Anthropic cannot prove intent or real-world use.
Anthropic banned users and improved AI safeguards based on these findings.
Claude AI was used by Iranian state-aligned accounts for influence operations and to develop targeting recommendations against U.S. naval forces.
AI was also used for mass surveillance platform design and transnational repression by actors in China and West Africa.
More advanced models like Claude Fable 5 have stronger safeguards and restrict dual-use biological research queries.
Anthropic noted that AI's biological capabilities have dual-use potential, similar to information used for weapons versus vaccines.

Continuer la lecture de l'article original