Coordinated Vulnerability Disclosure Timelines
- Source
- anthropic.com
- Author
- Anthropic research
- Date

Context
Anthropic, the AI company that makes Claude, will give software vendors and open-source maintainers much less time to fix the security flaws it reports to them before it makes those flaws public. Anthropic follows coordinated disclosure: it reports a vulnerability privately to whoever maintains the software and keeps it quiet until a deadline, called the disclosure window, so that a fix can ship before attackers learn of the flaw, while the deadline pushes the maintainer to act instead of leaving the report unanswered. That window has been 90 days. For vulnerabilities reported on or after January 1, 2027, Anthropic says it will cut it to 30 days for high- and critical-severity issues and 45 days for all others, while reports already in progress keep their existing deadlines. Two related deadlines move with it. When a maintainer does not answer the first report, Anthropic will aim to bring in an outside vulnerability coordinator after 15 days instead of 30, and once a patch is available it will wait 30 days instead of 45 before publishing full technical details.
Anthropic's reason is that the 90-day convention rested on two assumptions that no longer hold. Disclosure norms, its own included, were written when fixes were made at human speed and when it was unlikely that someone else would independently find the same bug during the waiting period. Models can now help write fixes, Anthropic says, and a bug that one model can find is increasingly likely to be found by another; it reports seeing more rediscovery of the same vulnerabilities as these capabilities improve. The second point is the one that changes what a disclosure window is worth. Keeping a flaw private protects users only while attackers do not know about it, and Anthropic says it no longer believes a vulnerability can be expected to stay private for months. Its own work is part of that change: it has described using an internal research model to find critical vulnerabilities in widely used browsers and operating systems (Discovering Vulnerabilities With Claude).
The shorter deadline has exceptions. For flaws in software that is hard to deploy, such as critical infrastructure, or flaws whose fix needs a change to a specification or RFC (the documents that define internet protocols), Anthropic will agree a remediation timeline with the vendor. Its seven-day window for vulnerabilities already being exploited stays the same. The new timelines also do not apply to AI-generated findings that no person has reviewed, such as those sent by OSS Scanner, Anthropic's free opt-in service that periodically scans critical open-source projects with its strongest models and sends the findings to their maintainers. Anthropic acknowledges that shorter timelines may put more pressure on maintainers, and says it will keep including actionable guidance and, where possible, a patch or mitigation with each report, and will agree on a sustainable pace with a project before sending it a large number of findings. It is also notifying maintainers of older high- and critical-severity reports that are past the 90-day deadline but still unpublished that they have 30 more days, after which Anthropic will publish those reports on its dashboard under the updated policy starting January 1, 2027.
Google's Project Zero, a team of Google security researchers who look for flaws in widely used software, reached the same conclusion by a different route. In a policy update dated November 13, 2026 (Intel), it called its old timeline clearly insufficient now that AI-assisted discovery and exploitation raise the risk that several parties, malicious ones included, find the same bug. Project Zero is moving gradually because vendors it consulted warned that an abrupt change could expose unfixed flaws: it discloses 90 days after reporting from that date, 60 days from February 2, 2027, and 30 days from January 3, 2028. It applies one deadline regardless of severity, arguing that severity changes a flaw's harm to users but not the chance that attackers find it, and it dropped its extra waiting period after a patch because attackers can now work out a bug from its patch almost immediately. Anthropic instead sorts by severity and keeps a waiting period after a patch, cut from 45 to 30 days. The two policies end in a similar place for maintainers: a serious flaw reported by Anthropic from January 1, 2027 has 30 days to be fixed before it becomes public, and Project Zero will reach a 30-day deadline for every flaw it reports a year later, because both organizations no longer trust a reported bug to stay secret for three months.
- Anthropic is cutting its disclosure deadline from 90 to 30 days for high- and critical-severity bugs and to 45 days for all others, for vulnerabilities reported on or after January 1, 2027. Reports already in progress keep their existing deadlines.
- Anthropic says the 90-day norm assumed fixes moved at human speed and that independent rediscovery was unlikely. It now sees a higher rediscovery rate, because a bug one model can find is increasingly likely to be found by another.
- Two related clocks also shorten: Anthropic will escalate to an external coordinator after 15 days of maintainer silence instead of 30, and will publish full technical details 30 days after a patch instead of 45.
- Exceptions remain: hard-to-deploy software such as critical infrastructure, and flaws needing spec or RFC changes, get negotiated timelines. The 7-day window for actively exploited bugs is unchanged, and unreviewed AI findings such as OSS Scanner output are excluded.
- High- and critical-severity reports already past the current 90-day deadline get a final 30-day extension, after which Anthropic will publish them on its dashboard under the updated policy.
Checking sign-in…
Loading comments…





