Anthropic will give vendors and open-source maintainers 30 days to fix high- and critical-severity vulnerabilities it reports before it discloses them, down from 90. All other bugs get 45 days. The change, announced by Anthropic's Frontier Red Team on November 16, applies to reports sent on or after January 1, 2027. Reports already in progress keep their current deadlines.
This matters because Anthropic's AI-driven bug hunting sends findings to many open-source projects, and a shorter clock puts the pressure on the people who have to triage and fix them. It arrives three days after Google Project Zero set its own shorter schedule, which we covered on Monday. Project Zero only reaches 30 days in January 2028.
What changes on January 1
| Current policy | From Jan 1, 2027 | |
|---|---|---|
| High/critical bugs: time before disclosure | 90 days | 30 days |
| All other bugs | 90 days | 45 days |
| Bugs under active exploitation | 7 days | 7 days (unchanged) |
| No reply from maintainer → escalate to outside coordinator | 30 days | 15 days |
| Full technical details after a patch ships | 45 days | 30 days |
All figures are from Anthropic's post. It says the rest of its disclosure policy is unchanged.
Backlog. High- and critical-severity bugs that Anthropic reported more than 90 days ago and that are still undisclosed get 30 more days. After that, Anthropic says it will publish them on its disclosure dashboard under the new rules from January 1. If you maintain a project that has an old, unresolved Anthropic report, this is the one to act on first.
Exceptions
- Hard-to-patch software. Anthropic names critical infrastructure and flaws that need a specification or RFC change. In those cases it says it will agree a timeline with the vendor. The post does not say what qualifies or how long an extension can run.
- Unreviewed AI findings. The new deadlines don't cover output that no human has reviewed, such as results from Anthropic's OSS Scanner service.
Why Anthropic says it's doing this
Anthropic's argument is about rediscovery. It says AI has made both finding and fixing bugs faster, and that a flaw one model finds is increasingly likely to be found by another. In its words, a 90-day window "made sense when a vulnerability could be expected to stay private for months." The Next Web reports the follow-on line: Anthropic no longer believes that.
The post doesn't give numbers to back the rediscovery claim. Neither TNW nor LavX News quotes any maintainers or outside experts. Both note that the shorter windows will add pressure on maintainers. Whether Anthropic's reports are worth that pressure is already disputed. VulnCheck's audit of its Glasswing findings argued that maintainers are drowning in low-value reports, as we reported last week.
What Anthropic offers in return
- Each report comes with guidance on fixing the bug and, where possible, a patch or mitigation.
- It says it will agree on a manageable pace with a project before sending it a large batch of findings.
- The Claude for Open Source Program offers six months of Claude Ultra at no cost to eligible maintainers and contributors. OSS Scanner offers free scanning for critical open-source projects.
What maintainers and vendors should do before January
- Clear old Anthropic reports. Anything high or critical that is more than 90 days old gets a 30-day grace period, then goes public.
- Publish a working security contact. After 15 days without a reply, Anthropic now goes to an outside coordinator. A security policy file, or GitHub private vulnerability reporting, gets reports to the right person.
- Plan for 30-day fixes. Check whether your release process can get a fix for a serious bug out within a month. If it can't, for example because downstream distributions need time, ask Anthropic for an extension early. Its post says it will negotiate for hard-to-deploy software.
- Ask for a pace agreement. If your project is likely to get many findings, use Anthropic's offer to agree on a pace before a batch arrives.
- Downstream users: expect public details on more bugs, sooner. Patch faster, especially for libraries you bundle and don't update automatically.
What to watch
Two of the most active bug-finding teams have now committed to 30-day windows: Anthropic from January 2027 and Project Zero from January 2028. Watch whether other AI labs and bug-bounty platforms follow, and whether Anthropic publishes data on how often its findings are found again by others. That data would test the case it is making for the change.