AI News
All stories
Official

Anthropic shortens its vulnerability disclosure window to 30 or 45 days

Anthropic is cutting the disclosure window in its coordinated vulnerability disclosure policy from 90 days to 30 or 45, citing faster AI-driven vulnerability discovery.

Safety & Security · · Updated · 2 min read

Anthropic is shortening the disclosure window in its coordinated vulnerability disclosure policy from 90 days to 30 days for high- and critical-severity issues, and to 45 days for everything else. The company says frontier models have become much better at finding vulnerabilities, and that the pace of AI progress has changed how it views its own disclosure practices. The new timelines apply to vulnerabilities reported to vendors and maintainers on or after January 1, 2027. Reports already in progress on that date keep their existing deadlines.

Anthropic's post argues that disclosure norms, including its own 90-day policy, rest on two assumptions that no longer hold. The first is that fixes move at human speed. The second is that independent rediscovery of the same bug is unlikely. According to the post, models can now help write fixes, and a bug one model finds is increasingly likely to be found by another. The company says it is seeing more rediscovery as capabilities rise. A 90-day window made sense when a flaw could be expected to stay private for months, the post says, but Anthropic no longer believes that is true.

Two related changes accompany the shorter window. Anthropic will aim to escalate to an external vulnerability coordinator after 15 days instead of 30 if a maintainer does not respond to its initial report. It will also wait 30 days instead of 45 after a patch is available before sharing full technical details. The post says the rest of the policy is unchanged.

The post also covers reports that are already overdue. High- and critical-severity flaws that Anthropic reported more than 90 days ago and that remain undisclosed will give maintainers an additional 30 days. After that, they will be published on Anthropic's dashboard under the updated policy starting January 1, 2027.

Some limits are built into the new timelines. According to the post, flaws in hard-to-deploy software, such as critical infrastructure, and flaws that require changes to a specification or RFC can't always be resolved in 30 or 45 days. In those cases, Anthropic will work with the vendor on a remediation timeline. The seven-day window for vulnerabilities under active exploitation is unchanged. The policy also does not apply to AI-discovered findings that have not received human review, such as those found by OSS Scanner. OSS Scanner is Anthropic's no-cost scanning service for critical open-source projects.

The post acknowledges that shorter timelines put more pressure on maintainers. Anthropic says it will keep including actionable guidance with each report, along with a patch or mitigation where possible. It will also agree on a sustainable pace with maintainers before sending a large volume of findings to a single project. The post points to OSS Scanner, the Claude for Open Source Program, which offers six months of Claude Ultra at no cost to eligible maintainers and contributors, and the Cyber Verification Program for security professionals as ways to help maintainers and defenders keep up.

What changed

Anthropic shortened its coordinated vulnerability disclosure window from 90 days to 30 days for high- and critical-severity issues and to 45 days for everything else, and cut the escalation and post-patch waiting periods.

Type: Other · Takes effect: Friday, January 1, 2027

Why it matters

Companies that depend on open-source or vendor software will face shorter windows to patch vulnerabilities that AI tools surface. That raises pressure on maintainers and on internal patch processes.

What you can do

Security teams that depend on Anthropic-reported findings should plan for 30- or 45-day disclosure windows for reports received from January 1, 2027.

How the story developed

  1. This update adds that the 30-day window applies to high- and critical-severity vulnerabilities.

Original source

Where we saw it

2 sources

Anthropic Research, Latent Space

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works