Anthropic shortens bug disclosure window to 30 days
The AI company is accelerating its security reporting timeline because AI models can now identify software vulnerabilities much faster than humans.

The short version
- Serious and high-severity flaws will move from a 90-day disclosure window to 30 days.
- The new policy takes effect for flaws reported on or after January 1, 2027.
- Anthropic notes that AI models find bugs faster, making it more likely for multiple models to find the same flaw.
- Google's Project Zero is also planning to shorten its disclosure windows to 60 and then 30 days.
- Flaws under active attack will maintain a 7-day disclosure window.
Why it matters
This change signals a shift in the cybersecurity landscape where the speed of automated discovery outpaces traditional human-led patching cycles. For software maintainers, it means a much tighter window to secure their code before vulnerabilities become public knowledge.
Accelerating the security cycle
Anthropic has announced a significant change to how it handles security vulnerability disclosures. The company's Frontier Red Team will now publish serious and high-severity flaws just 30 days after they are reported to software maintainers. This is a sharp reduction from the previous 90-day window that the company used to provide for fixes.
The change is driven by the increasing capabilities of artificial intelligence in identifying software bugs. Anthropic argues that because AI models can find flaws so quickly, the old timeline of several months is no longer realistic. The company believes that a bug discovered by one model is increasingly likely to be discovered by another model shortly thereafter.
As reported by The Next Web, these new rules will officially apply to any flaws reported on or after January 1, 2027. However, the company is already addressing older flaws that have exceeded their current deadlines. Anthropic is granting maintainers of these specific older flaws an additional 30 days before the reports are made public on its dashboard.
New rules for maintainers
Beyond the primary 30-day window for serious flaws, Anthropic is adjusting several other procedural steps for security reporting. All flaws that do not meet the high or critical severity threshold will now be given a 45-day window for resolution. This ensures that even less severe issues are addressed with a standardized timeline before they are shared with the public.
The company is also speeding up its communication with software maintainers who might be unresponsive. If a maintainer fails to answer an initial report, Anthropic will move the issue to an outside coordinator after 15 days, which is a 15-day reduction from the previous 30-day wait. These changes are designed to keep the flow of information moving more efficiently.
Once a software patch is actually released, Anthropic will wait 30 days before sharing the full technical details of the flaw. This is a reduction from the previous 45-day waiting period. These adjustments collectively aim to minimize the time a vulnerability remains known to the company but hidden from the broader security community.
Exceptions for critical systems
Anthropic recognizes that some types of software require much longer periods to secure than others. For instance, flaws found in critical infrastructure software or within technical standards may take more than 30 or 45 days to properly remediate. In these specific instances, Anthropic says it will work directly with the vendor to agree on a custom timeline for disclosure.
The company is also maintaining a very aggressive 7-day disclosure window for any flaws that are currently under active attack. This high-priority category ensures that immediate threats are handled with the utmost urgency. These exceptions provide a safety valve for complex systems where a rushed fix might cause more harm than good.
It is important to note that these new disclosure policies do not apply to all AI-generated findings. Specifically, the policy does not cover findings from the OSS Scanner service that have not yet been reviewed by a human. Anthropic continues to prioritize human oversight in the verification of security flaws before they enter the reporting pipeline.
A shift in the industry
Anthropic's decision reflects a broader trend in the cybersecurity industry toward faster disclosure cycles. The company noted that the old 90-day rules were based on the assumption that fixes moved at human speed and that two people rarely found the same bug. They no longer believe that these assumptions hold true in an era of automated discovery.
Other major players in the security space are moving in the same direction. Google’s Project Zero is also planning to shorten its disclosure windows. According to SecurityWeek, Project Zero will begin publishing flaws 60 days after reporting them starting in February 2027, with a further drop to 30 days scheduled for January 2028.
Anthropic acknowledges that these shorter deadlines may put significant pressure on the people who maintain open-source and commercial software. To help mitigate this, the company stated it will continue to provide fix guidance and, whenever possible, provide a patch alongside each report. The goal is to provide maintainers with the tools they need to act quickly under a tighter schedule.
The future of bug hunting
The move to a 30-day window marks a transition toward a more proactive and rapid security posture. By shortening the time between discovery and public disclosure, Anthropic aims to reduce the window of opportunity for malicious actors to exploit a known flaw. This approach forces a faster reaction from the software development community as a whole.
For software developers, this means that the
Sources and links
- OfficialFrontier Red Team said on Monday anthropic.com
- Officialsecurityweek.com/google-project-zero-sets-course-for-30-day-disclosure securityweek.com
- ReportingThe Next Web: the article we sourced
First reported by The Next Web. 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.