Anthropic Revamps Vulnerability Disclosure Timeline, Reducing Timeframes for Major Flaws
In a significant shift in its vulnerability disclosure strategy, Anthropic has decided to shorten the timeframe vendors have to address serious flaws identified by its Claude models. Originally set at 90 days, this period will now be reduced to just 30 days. This policy alteration specifically pertains to vulnerabilities that Anthropic discovers and reports to vendors and software maintainers, rather than addressing bugs reported by external researchers.
Announced by the company’s Frontier Red Team in a post dated November 16, the new timelines will come into effect for all vulnerabilities reported on or after January 1, 2027. It is important to note that reports which are currently being processed will retain their existing deadlines and will not be subjected to this new policy.
Anthropic’s decision stems from a recognition that traditional disclosure norms—particularly assumptions around the pace of fixes—have become outdated. The company acknowledges that past expectations considered the resolution of vulnerabilities to operate at a human speed, assuming that independent rediscovery of issues would be rare. However, Anthropic asserts that these assumptions no longer hold true, as models are becoming more efficient and capable of discovering vulnerabilities that may have already been identified by other systems.
Key Changes in the Disclosure Policy
Under the revised policy, vulnerabilities classified as high- or critical-severity will now have a tighter 30-day disclosure deadline. For all other issues, the deadline will extend to 45 days. In addition, the company has made commitments to escalate issues to an external vulnerability coordinator after a mere 15 days if there is no response from the maintainer, a notable reduction from the prior 30-day period.
Moreover, Anthropic will withhold complete technical details for 30 days following the release of a patch, instead of the previous 45 days. The disclosure window for vulnerabilities that are actively being exploited will remain unchanged at seven days.
Anthropic has also indicated its commitment to collaborate with vendors on remediation plans for vulnerabilities associated with software that is challenging to deploy—particularly critical infrastructure—and for issues that require specification or RFC changes. However, findings generated by AI models that have not undergone human review, such as those identified through the Open Source Software (OSS) Scanner, will not fall under this policy.
While industry professionals have raised concerns that these shortened timeframes may exert additional pressure on maintainers, Anthropic reassures that they will continue to provide actionable guidance with each vulnerability report and strive to include a patch or mitigation where feasible.
Comparison with Project Zero’s Policy
Anthropic’s new vulnerability deadlines now precede those established by Google’s Project Zero, which currently publishes details about flaws 90 days after reporting, regardless of whether they have been addressed. However, Project Zero plans to update its timeline to 60 days on February 2, 2027, and subsequently to 30 days on January 3, 2028.
Interestingly, the two organizations diverge in their approach to categorizing vulnerability severity. Anthropic employs different disclosure windows based on severity levels of the vulnerabilities. In contrast, Project Zero maintains that a vulnerability’s seriousness does not necessarily correlate with how likely attackers are to discover it independently.
Coincidentally, on the same day as Anthropic’s announcement, Google’s Threat Intelligence Group released findings regarding an exposed server running an attack framework known as "exploit_agent," which is designed for vulnerability research and weaponization. This revelation adds another layer of urgency to the discourse on vulnerability disclosures and highlights the challenges and risks in maintaining security in an increasingly complex digital landscape.
In response to the competitive and evolving landscape of cybersecurity, Anthropic is also granting vendors an additional 30 days to address high- and critical-severity bugs that have exceeded the previous 90-day deadline before making them public on its disclosure dashboard. This decision strikes a balance between transparency and the necessity for responsible remediation.
As Anthropic continues to refine its approach to vulnerability disclosures, stakeholders across the tech industry will be watching closely, eager to see how these changes will impact software security and the overall landscape of cybersecurity practices.

