Anthropic Cuts Vulnerability Disclosure Deadlines to 30 Days From January 2027
Anthropic will give software makers just 30 days to fix serious security flaws its AI models find, down from 90 days, starting 1 January 2027 . Less serious flaws get 45 days. The reason is blunt: a bug one AI model can find is increasingly likely to be found by another, so Anthropic no longer believes a vulnerability can be expected to stay private for months. Coordinated disclosure is the practice of reporting a security flaw privately and giving the vendor a fixed window to fix it before it is published. Anthropic announced the shorter deadlines on 16 November 2026 in a policy update from its Frontier Red Team, titled "Updating our coordinated vulnerability disclosure policy timelines". The change applies to vulnerabilities its models report to vendors and open-source maintainers on or after 1 January 2027. What exactly changes The headline number is 30 days, but four separate clocks are moving, and they matter in different ways depending on which side of a report y...