DATUM

Anthropic says five attempts to use Claude for bioweapons work were made

single source · 1 articles · society · confidence: low · first seen 2026-09-11 16:34 UTC

Anthropic says it found five cases in which people tried to use its Claude models to support biological weapons development, according to The Guardian. The company banned the accounts. In one case, a platform sent requests that Claude had rejected to a rival with weaker safeguards. The same editorial reports the US and China will hold their first bilateral AI-safety talks before a planned Trump-Xi summit. It notes an Anthropic researcher has warned of possible extinction risk. The piece argues AI safety cannot depend on the least responsible model. No official confirmation of the talks was given.

What this means for you

If you run an AI platform, account bans are not enough: one platform routed Claude-rejected requests to a weaker rival. Safety teams should assume misuse attempts will move to the least protected model. Policy teams should watch for confirmation of the US-China AI-safety talks; nothing binding exists yet.

Key facts

  • ·Anthropic says it identified five cases where attempts were made using its Claude models to support biological weapons development. source
  • ·Anthropic banned the accounts involved. source
  • ·In one case, a platform sent requests rejected by Claude to a rival with weaker safeguards. source
  • ·The US and China will reportedly hold their first bilateral AI-safety talks before a planned Trump-Xi summit. source

What the sources say

  • The Guardian AI (press) — Opinion piece using Anthropic's bioweapons misuse report to argue for government AI safety coordination.

Sources

The original reporting. Follow these — they did the work.

← the wire