Anthropic has announced a significant shift in its vulnerability disclosure policy, reducing the time frame vendors are granted to address security flaws before the company makes them public. Effective January 1, 2027, the company’s Frontier Red Team will enforce a 30-day disclosure deadline for vulnerabilities classified as high or critical severity. This update represents a substantial acceleration from the previous 90-day standard.
The policy change applies specifically to vulnerabilities identified by Anthropic’s own models and reported to external vendors or maintainers. It does not extend to general bug reports submitted to the company by independent researchers. According to the company, the adjustment is a response to the evolving speed of the threat landscape, noting that traditional disclosure norms were built on the assumption that fixes moved at human speed and that independent rediscovery of bugs was unlikely. Anthropic stated that these assumptions are no longer valid, as the capabilities of modern AI models have increased the likelihood that a vulnerability discovered by one system will soon be identified by another.
Under the new guidelines, vulnerabilities categorized as high or critical will face a 30-day deadline, while lower-severity issues will be subject to a 45-day window. The company has also tightened its escalation procedures; if a maintainer fails to respond within 15 days, Anthropic intends to involve an external vulnerability coordinator, cutting the previous 30-day escalation wait time in half. Furthermore, the period Anthropic waits to publish full technical details after a patch is released will be reduced from 45 days to 30 days.
The company confirmed that certain exceptions will remain in place. The seven-day disclosure window for vulnerabilities currently under active exploitation will not change. Additionally, Anthropic stated it would collaborate with vendors to establish flexible remediation timelines for complex scenarios, such as flaws affecting critical infrastructure or those requiring significant specification or RFC changes. The policy also excludes AI-generated findings that have not undergone human review, such as those produced by its OSS Scanner.
To mitigate the increased pressure on software maintainers, Anthropic committed to providing actionable guidance alongside each report, including potential patches or mitigations whenever possible. For reports currently in progress by the start of the new year, the previous deadlines will remain in effect. Additionally, for high and critical bugs that have already surpassed the current 90-day threshold, the company will grant an extra 30 days before publishing the details on its disclosure dashboard.
This policy update aligns Anthropic with a broader industry trend toward faster disclosure cycles. For comparison, Google’s Project Zero is also moving toward shorter timelines, with plans to cut its disclosure window to 60 days by February 2027 and to 30 days by January 2028. However, the two organizations maintain different philosophies regarding severity; while Anthropic intends to adjust its windows based on the risk level of the vulnerability, Project Zero has maintained that the severity of a bug does not necessarily correlate with the speed at which an attacker might discover it.




