Anthropic to shorten vulnerability disclosure window to 30 days
The AI company is accelerating its security timeline because AI models make it easier for hackers to find and replicate software bugs.

The short version
- Serious security bugs will now be publicly disclosed after 30 days, down from the previous 90-day window.
- Minor bugs will move from a 90-day disclosure window to 45 days.
- Anthropic will now escalate reports to third-party coordinators after 15 days if a developer fails to respond.
- The new policy takes effect for all reports submitted on or after January 1, 2027.
- The change acknowledges that AI models can find vulnerabilities much faster than human researchers alone.
Why it matters
This shift signals that the 'human speed' of software patching is no longer sufficient to keep up with AI-driven security threats. For users, it means faster public warnings about flaws, while for developers, it creates a much tighter window to secure their code.
A faster public warning system
Anthropic is updating its coordinated vulnerability disclosure policy to significantly shorten the time between reporting a bug and making it public. Starting January 1, 2027, the company will disclose high- and critical-severity vulnerabilities after 30 days instead of the current 90-day period. This change aims to ensure that serious security flaws do not remain hidden for months while they are being addressed.
The update also affects less severe bugs, which will move to a 45-day disclosure window. For reports that are already open before the 2027 deadline, the original timelines will remain in place. However, Anthropic is already beginning to notify maintainers of older, overdue high-severity reports that they will publish within 30 days.
The company plans to publish these details on its own dashboard once the disclosure period ends. This move provides a more consistent timeline for security researchers and the public to see how vulnerabilities are being handled. By shortening these windows, Anthropic intends to reduce the period during which a flaw might be exploited without public knowledge.
Why the timeline is shrinking
Anthropic stated that its previous policies were written before AI could be used to find software vulnerabilities. The company notes that old policies assumed fixes moved at human speed and that it was unlikely for independent researchers to find the same bug. Neither of those assumptions holds true in the current technological landscape.
As AI capabilities grow, models are increasingly able to find the same vulnerabilities across different systems. Anthropic observes that models can also help write the actual fixes for these flaws. Because the discovery of bugs is accelerating, the company believes that keeping a reported vulnerability private for a long time is no longer practical or safe.
The new rules reflect a shift toward a faster-paced security environment where automated tools can rapidly scan for weaknesses. Anthropic argues that because a bug found in one model is likely to be found by another, the window for private remediation must be smaller. This adjustment ensures that the speed of disclosure matches the speed of modern discovery.
Impact on software maintainers
The new policy places significantly more pressure on the people who maintain open-source software. For serious flaws, the time available to triage, fix, and ship a patch is being cut by two thirds. To help manage this, Anthropic will continue to provide actionable guidance and, when possible, a specific patch or mitigation with every report.
Anthropic also plans to coordinate with maintainers to ensure they are not overwhelmed by a large batch of findings at once. They will agree on a workable rate for sending reports to large projects. This is intended to help maintainers stay on top of the faster timeline without being flooded by too many requests simultaneously.
For those who do not respond to the initial report, the escalation process is also speeding up. Anthropic will now aim to contact an external vulnerability coordinator after 15 days, down from 30 days. These third parties act as brokers to facilitate contact between the person who found the bug and the vendor who needs to fix it.
Exceptions and specific cases
Not every type of bug will fall under the new 30-day disclosure rule. Anthropic noted that flaws in critical infrastructure or software that is difficult to deploy may require a different approach. In these cases, the company will work directly with the vendor to establish a custom remediation timeline based on the complexity of the fix.
The policy also excludes findings that have not yet undergone human review. This specifically includes data from OSS Scanner, which is a free service Anthropic provides to scan the code of critical open-source projects. Anthropic has not yet specified what disclosure timeline applies to these specific automated findings.
The company will also wait 30 days after a patch is officially released before publishing full technical details. This is a reduction from the previous 45-day wait. These layers of timing are designed to ensure that a fix is actually available to the public before the specific details of the vulnerability are revealed.
What to watch
- How many maintainers struggle to meet the new 30-day deadline.
- Whether Anthropic clarifies the disclosure timeline for findings from the OSS Scanner.
- How other AI companies adjust their disclosure policies in response to Anthropic's move.
Try it
If you manage open-source projects, check if you are eligible for the Claude for Open Source Program to receive six months of Claude Ultra for free.
Sources and links
- Officialanthropic.com/research/coordinated-vulnerability-disclosure-timelines anthropic.com
- Reportinghelpnetsecurity.com: the article we sourced
First reported by helpnetsecurity.com. 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.