SIAgentNews.com
← All news

Anthropic shortens vulnerability disclosure windows due to AI speed

The AI company is cutting the time vendors have to fix security flaws from 90 days to 30 days for high-severity issues as automated discovery accelerates.

· First reported by Anthropic

30 daysNew disclosure window for high-severity issues

The short version

  • High-severity security flaws must now be disclosed within 30 days instead of the previous 90-day window.
  • Standard vulnerabilities will have a 45-day disclosure window starting January 1, 2027.
  • Anthropic will escalate unresponsive reports to external coordinators after 15 days instead of 30.
  • The company is offering free scanning and Claude Ultra access to help maintainers manage the faster pace.

Why it matters

This marks a shift in how the tech industry handles security, acknowledging that AI makes finding bugs much faster than humans alone. For users, it means faster public awareness of flaws, while for developers, it means a tighter deadline to secure their code.

The impact of AI on security

Anthropic recently announced a significant shift in its coordinated vulnerability disclosure policy. The company is shortening the timeframes allowed for software developers to fix security bugs before the details are made public. These changes are a direct response to the fact that frontier AI models have become much more proficient at discovering vulnerabilities.

Historically, security disclosure norms were built on the assumption that humans were the primary actors in finding bugs. In that era, it was believed that fixes moved at human speeds and that it was unlikely for an independent party to find the same bug quickly. Those assumptions no longer hold true in an environment where AI can scan and identify flaws at scale.

The company notes that because models can now help develop fixes, a bug found by one model is increasingly likely to be found by another. This leads to a higher rate of vulnerability rediscovery. To keep pace with these capabilities, Anthropic believes the previous 90-day window is no longer sufficient to keep vulnerabilities private.

New disclosure timelines

Starting January 1, 2027, the new policy will establish much tighter deadlines for reporting security flaws. High-severity and critical-severity issues will move from a 90-day disclosure window to just 30 days. For all other types of vulnerabilities, the disclosure window will be reduced to 45 days.

The company is also changing how it handles communication with software maintainers. If a maintainer does not respond to an initial report, Anthropic will now aim to escalate the issue to an external vulnerability coordinator after 15 days. This is a reduction from the previous 30-day waiting period for a response.

Additionally, the company will wait 30 days after a patch becomes available before sharing full technical details with the public. This is a reduction from the previous 45-day wait. However, the company will maintain its existing seven-day disclosure window for any vulnerabilities that are currently being exploited in the wild.

Exceptions and grace periods

Anthropic acknowledges that some types of software cannot be patched within these new, shorter windows. For example, flaws in critical infrastructure or software that requires changes to a formal specification or RFC may take longer to resolve. In these specific instances, the company will work directly with the vendor to establish a custom remediation timeline.

The company is also addressing a backlog of high-severity vulnerabilities that were reported over 90 days ago. Because these reports exceed the current policy deadlines but remain undisclosed, Anthropic is notifying the maintainers that they have an additional 30 days to fix them. After that period, the reports will be published on the company's dashboard.

It is important to note that these new rules do not apply to AI-discovered findings that have not yet been reviewed by a human. For example, findings from the OSS Scanner tool are excluded from these specific disclosure timelines. The policy is intended to govern reports that have reached a level of human verification.

Support for maintainers

The company recognizes that shorter deadlines place significant pressure on the people who maintain open-source software. To help manage this burden, Anthropic will continue to provide actionable guidance with every report. They aim to include a patch or a specific mitigation whenever possible to help maintainers confirm and fix the issue quickly.

To prevent overwhelming any single project, Anthropic will continue to agree on a sustainable pace with maintainers before sending large volumes of findings. They are also offering several programs to help defenders keep up with the faster pace of AI-enabled discovery. These include a no-cost service that periodically scans code with their strongest models.

Maintainers of critical open-source projects can apply to the OSS Scanner service to receive direct findings. Additionally, eligible maintainers can apply for six months of Claude Ultra at no cost. Security professionals can also apply to the Cyber Verification Program, which tailors the model's safeguards to support legitimate defensive work.

What to watch

  • How other AI companies adjust their disclosure policies in response to automated bug discovery.
  • Whether the 30-day window leads to a decrease in the quality of patches due to time pressure.

Try it

Check the Anthropic website to see if your open-source project is eligible for the OSS Scanner service or the Claude for Open Source Program.

Sources and links

First reported by Anthropic. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news