Anthropic has turned on code scanning with its strongest cybersecurity model, Mythos 5, inside Claude Security and in partner defensive products, giving users the findings and suggested fixes rather than access to the model itself. Scans return flaws tagged by category, severity and confidence, and every fix must be approved by a person before it is applied. The company is also putting $35mn of usage credits into a fund for open-source security work, three weeks before EU vulnerability reporting duties begin.
What changed
Mythos 5 was held back from broad access, with security teams unable to use it for code scanning.
What it unlocks
Enterprise customers can scan a code repository with Anthropic's strongest security model and get ranked findings with suggested fixes, billed as normal usage.
- $35mn in credits for open-source security
- CRA reporting starts 11 September
- 10,000 critical flaws found in a month
What you need to act on it
- Anthropic enterprise account
- human review and approval of every patch
- access via Claude Security or a partner product rather than the model itself
Sources