Infosecurity
Anthropic cuts serious vulnerability disclosure deadline to 30 days
Anthropic has shortened the disclosure deadlines for serious flaws found by its Claude models, cutting the window from 90 days to 30. The change, announced by Anthropic’s Frontier Red Team in a 16 November post, applies to vulnerabilities Anthropic reports to vendors and maintainers and to reports submitted on or after 1 January 2027; issues already in progress will retain their existing deadlines.
The policy shift reflects concerns that bugs found by one model are increasingly likely to be found by others and that fixes are not moving at “human speed.”
Under the updated policy, high- and critical-severity vulnerabilities will have a 30-day disclosure deadline, while all other issues will be disclosed within 45 days. If a maintainer does not respond within 15 days, Anthropic will escalate to an external vulnerability coordinator (instead of waiting 30 days). In addition, full technical details will be published 30 days after a patch becomes available, down from 45 days.
The seven-day disclosure window for vulnerabilities under active exploitation remains unchanged. The policy notes that it will work with vendors on remediation timelines for hard-to-deploy software or needs for specification changes, and it will not cover AI-discovered findings that have not received human review, such as those from its OSS Scanner.
The article also contrasts these timelines with Project Zero, which will tighten its disclosures to 60 days in February 2027 and 30 days in January 2028, and notes Google's Threat Intelligence Group recently found an server hosting an attacker framework called exploit_agent.