Anthropic Shrinks Vulnerability Deadlines Amid AI Bug Surge

Anthropic has shortened the disclosure deadlines for security flaws, cutting the timeline for high- and critical-severity vulnerabilities from 90 to 30 days and giving vendors 45 days for everything else. This change reflects how fast AI tools can now spot flaws.

What’s New in the Policy

The updated policy applies to vulnerability reports submitted on or after January 1, 2027. Issues already under review before that date will stick with the previous deadlines. For reports with high or critical severity, Anthropic will also reduce the coordination window—from 30 to 15 days—and will publish full technical details 30 days after a patch is released, instead of the earlier 45-day wait.

Some older reports are affected too. For vulnerabilities reported more than 90 days ago that haven’t yet crossed their disclosure deadline, maintainers now have 30 additional days to remediate before Anthropic publishes the findings under the new rules.

Why the Change Matters

Anthropic argues that the prior 90-day standard was based on an era before AI-driven bug discovery when patches took longer and redundant reports were rare. Now, bugs found by one model often get rediscovered by another, which increases the chance that a flaw made public after months is already circulating secretly.

The scale of discoveries supports the change. Since November 2, 2026, over 5,000 vulnerabilities across 680 open-source projects have been disclosed via Anthropic’s dashboard, but only 629 had confirmed patches. Internal tools generated around 29,000 candidate flaws, though only about 6,000 have been manually reviewed. These bottlenecks in human review have stretched remediation timelines.

Exceptions, Support, and What’s Expected

Anthropic acknowledges that in some sectors—especially critical infrastructure—faster fixes may be unrealistic. Altering a specification or changing an RFC could require more time. In those cases, tailored timelines will be negotiated. Flaws under active attack still get a 7-day deadline. The policy does not apply to AI-generated findings that haven’t yet passed human review, such as outputs from its free OSS Scanner service.

Smaller open-source teams will feel additional pressure. To ease the transition, Anthropic plans to include detailed remediation steps in reports and offer workaround guidance where possible. Projects deemed critical can access Claude Ultra via its Open Source Program, and eligible maintainers can join its Cyber Verification Program, which includes enhanced safeguards for security-related work.

Developers and security teams relying on open-source code are advised to expect faster public disclosure and shorter patch windows. Practices like maintaining an accurate inventory of dependencies, staying tuned to Anthropic’s vulnerability dashboard, and preparing hotfix plans will become more critical than ever.

Analytical Take

This move signals a broader shift in vulnerability disclosure norms: AI has compressed the timeline for both detection and risk exposure. By tightening deadlines, Anthropic isn’t just adapting—it’s setting the pace for how industry standard policies must evolve. Key questions remain: how smaller contributors will cope, whether patch quality will suffer under shorter deadlines, and how similar frameworks will adapt across open-source ecosystems. The next year will test whether speed and security can go hand in hand.