Anthropic cuts its bug disclosure window from 90 days to 30
From 1 January, maintainers get 30 days for high and critical bugs Anthropic finds, 45 for the rest, down from a flat 90. The reason: AI models now rediscover the same flaws.
Anthropic is shortening the coordinated disclosure timelines it uses when its own security research, including the Frontier Red Team and OSS Scanner, reports bugs to maintainers. From 1 January 2027, high- and critical-severity issues get 30 days before Anthropic goes public, down from a flat 90. Everything else gets 45 days.
The knock-on deadlines move too: once a patch ships, Anthropic now waits 30 days instead of 45 before publishing full technical detail, and escalates to an external coordinator after 15 days instead of 30 if a maintainer stays silent. The seven-day window for bugs already under active exploitation is unchanged. Anthropic says the old 90-day figure “made sense when a vulnerability could be expected to stay private for months”, and that frontier models are now good enough at finding bugs that the same flaw is likely to be rediscovered by a different AI system while the first report is still sitting quietly with a maintainer.
Why it matters: if your project gets a report from Anthropic’s OSS Scanner or Cyber Verification Program, you now have a third of the runway you had before to patch ahead of public disclosure. Anthropic is pointing maintainers at that same OSS Scanner, six months of free Claude for Open Source access, and the Cyber Verification Program to help absorb the shorter clock.
The caveat: the new timelines don’t cover AI-discovered findings nobody has reviewed yet, such as raw OSS Scanner output, and fixes needing critical-infrastructure coordination or a spec change still get a negotiated schedule instead.