Anthropic, the company behind the Claude AI models, said on Monday, 16 November, that it will give software makers far less time to fix security flaws that it reports to them before it makes those flaws public. The window for high- and critical-severity vulnerabilities will drop from 90 days to 30 days, and to 45 days for all other issues.

The change was announced by the company's Frontier Red Team in a post on Anthropic's website. It applies to vulnerabilities reported on or after 1 January 2027; reports already in progress will keep their existing deadlines.

Why the window is shrinking

Coordinated vulnerability disclosure is the practice under which a researcher who finds a security bug tells the software's maintainer privately and waits for a set period before publishing details, so that a fix can be prepared.

Anthropic said advances in AI have changed how it views that practice. Frontier models have become much better at finding vulnerabilities, the company said, and older disclosure norms rested on two assumptions: that fixes move at human speed and that someone else is unlikely to find the same bug independently. It said neither holds any longer and that it is seeing more cases of the same flaw being rediscovered.

A 90-day window "made sense when a vulnerability could be expected to stay private for months," the post said, adding that the company no longer believes that to be true. The post does not give figures on how often flaws are being rediscovered or on how long patches now take.

What else changes

  • If a maintainer does not respond, Anthropic will aim to escalate to an external vulnerability coordinator after 15 days instead of 30.
  • Full technical details will be shared 30 days after a patch becomes available, down from 45 days.
  • The seven-day window for vulnerabilities that are already being actively exploited stays as it is.
  • The rest of the policy is unchanged.

For its backlog, the company said high- and critical-severity flaws that were reported more than 90 days ago, are past the current deadline and are still undisclosed will get an additional 30 days. After that they will be published on Anthropic's dashboard under the updated policy, starting 1 January 2027.

Exceptions and help for maintainers

Anthropic acknowledged that some software cannot realistically be fixed in 30 or 45 days. It cited systems that are hard to update, such as critical infrastructure, and flaws that require changes to a specification or an RFC, the documents that define internet standards. In those cases it will agree a remediation timeline with the vendor.

The policy does not cover findings made by AI that have not been reviewed by a human, such as those from its OSS Scanner, the post said.

To reduce the burden on those receiving reports, the company said its reports will carry actionable guidance and, where possible, a patch or mitigation, and that it will agree on a sustainable pace with maintainers before sending a large number of findings to a single project. It pointed to its OSS Scanner, which offers no-cost scanning for critical open-source projects, and to a programme that gives eligible open-source maintainers and contributors six months of its Claude Ultra plan free.

What happens next

The shorter deadlines take effect for reports filed from 1 January 2027. Maintainers with older, overdue high-severity reports from Anthropic have the additional 30 days before those begin appearing on the company's public dashboard.

PS