Anthropic’s Frontier Red Team announced on 16 November 2026 that it is slashing its coordinated vulnerability disclosure window from 90 days to 30 days for high- and critical-severity issues. This is not a minor policy tweak; it is an admission that the foundational assumption of the 90-day system—that a bug can remain private for months—has collapsed under the weight of AI-driven discovery. The new timelines take effect for reports made on or after 1 January 2027.

The Mechanics of the New Clock

The change involves four distinct moving parts that will reshape how maintainers interact with security researchers. For high- and critical-severity issues, the window drops to 30 days, while all other issues move to a 45-day window. Crucially, the escalation clock for unresponsive maintainers has been halved: if a maintainer does not respond to the initial report, Anthropic will now escalate to an external vulnerability coordinator after just 15 days. Furthermore, full technical details will be published 30 days after a patch ships, down from 45 days.

Why the 90-Day Bet Failed

Anthropic’s reasoning is blunt: disclosure norms were built when fixes moved at human speed and independent rediscovery was unlikely. Neither assumption holds in the current landscape. As frontier models improve, the likelihood that a flaw found by one model is rediscovered by another increases dramatically. Anthropic explicitly cites its 27 October 2026 report, where an unreleased Claude model built five real-world exploit chains across Chrome, Safari, Android, Windows, and Linux, and identified 62 Chrome CVEs. The window between "one lab knows" and "every lab knows" is shrinking, making the 90-day secrecy bet untenable.

Industry Consensus Forms

This move does not happen in a vacuum. Just three days prior, on 13 November 2026, Google Project Zero scrapped its long-standing "90 + 30" policy for a 30-day clock of its own. When two of the industry’s largest vulnerability-finding operations land on the same number within the same week, it signals a hard consensus. The 90-day era is not ending because of a single company’s impatience; it is ending because the ecosystem has collectively decided that AI-speed discovery requires AI-speed disclosure.

Support for Maintainers

Anthropic acknowledges the pressure this places on open-source maintainers and is offering concrete support to help them keep pace. The company is providing free access to its OSS Scanner, which uses Anthropic’s strongest models to scan critical projects. Eligible maintainers can also access six months of Claude Ultra for free through the Claude for Open Source Program. Additionally, the Cyber Verification Program will tailor Claude’s safeguards to enable legitimate defensive work for security professionals, ensuring that the tools used to find bugs can also help fix them.

Key Takeaways

  • Anthropic is reducing its coordinated vulnerability disclosure window to 30 days for high- and critical-severity issues, effective for reports submitted on or after 1 January 2027.
  • The escalation period for unresponsive maintainers is halved to 15 days, and the public release of technical details now occurs 30 days after a patch ships.
  • This shift aligns with Google Project Zero’s recent move to a 30-day clock, reflecting an industry consensus that AI-driven discovery rates have made 90-day secrecy windows obsolete.
  • Anthropic is introducing support mechanisms, including free OSS Scanner access and six months of Claude Ultra, to assist open-source maintainers in meeting the tighter deadlines.

The Bottom Line

The 90-day disclosure window was a human-speed artifact, and it is dead. If you maintain software, your new reality is a 30-day clock backed by models that can find bugs faster than you can patch them.