Anthropic Cuts Its Vulnerability Disclosure Window to 30 Days as AI Speeds Up Bug Finding
Anthropic has announced a shorter disclosure window in its coordinated vulnerability disclosure (CVD) policy. For high- and critical-severity issues, the window drops from 90 days to 30 days. For all other issues it falls to 45 days. The new timelines apply to vulnerabilities reported to vendors and maintainers on or after 1 January 2027, and in-progress reports keep their existing deadlines. According to Infosecurity Magazine, the change covers bugs Anthropic reports to vendors and maintainers, not generally those reported to Anthropic by outside researchers, and was announced by the company's Frontier Red Team in a November 16 post.
Two related changes accompany the shift. Anthropic will aim to escalate to an external vulnerability coordinator after 15 days, rather than 30, if a maintainer does not respond to an initial report. It will also wait 30 days instead of 45 after a patch is available before sharing full technical details. Some high- and critical-severity reports already older than 90 days will receive an extra 30 days, then be published on Anthropic's dashboard in line with the updated policy starting 1 January 2027.
The company's reasoning is that existing disclosure norms were built before AI could find vulnerabilities. Those norms assumed fixes moved at human speed and that independent rediscovery of the same bug was unlikely. Infosecurity Magazine quotes Anthropic as saying "neither assumption holds today": models can help develop fixes, and a bug one model finds is increasingly likely to be found by another. Anthropic reports seeing a higher rate of rediscovery as capabilities grow.
There are exceptions. Flaws in hard-to-deploy software, such as critical infrastructure, or those needing changes to a specification or RFC, will get a remediation timeline agreed with the vendor. The seven-day window for actively exploited vulnerabilities is unchanged, and the update does not cover AI-discovered findings without human review, such as those from its OSS Scanner. To ease the burden on maintainers, Anthropic will keep supplying actionable guidance and patches or mitigations where possible, and agree a sustainable pace before sending large volumes of findings. It also points to its Claude for Open Source Program and Cyber Verification Program.
Infosecurity Magazine adds context from elsewhere in the industry. It says the move puts Anthropic ahead of Google's Project Zero, which will cut its 90-day window to 60 days on February 2, 2027 and to 30 days on January 3, 2028. Project Zero argues severity does not change how likely attackers are to find a bug, while Anthropic sets windows by severity. On the same day, Google's Threat Intelligence Group reported an exposed server hosting an attacker framework called exploit_agent, built to research and weaponize vulnerabilities.
Why it matters
The long private window that many patching schedules quietly rely on is no longer a safe assumption, because Anthropic says AI makes independent rediscovery of the same bug increasingly likely. Organisations running open-source components or vendor software on a quarterly patch cycle are most exposed, since technical details may become public 30 days after a patch is available. Google's report of an exploit-building framework on an exposed server adds weight to the concern about faster weaponization.
What to do
- Review patch timelines for high- and critical-severity issues and shorten them where they exceed 30 days.
- Identify open-source components and vendor software that rely on quarterly update cycles.
- Track vendor and maintainer advisories so patches are applied soon after release, before full technical details are published.