When Anthropic’s Frontier Red Team finds a security hole in someone else’s software, it reports it to the vendor or maintainer and publishes it after a fixed deadline. That deadline is getting a lot shorter. For high- and critical-severity issues it drops from 90 days to 30, and for everything else to 45. Anthropic announced the change on November 16. It applies to vulnerabilities reported on or after January 1, 2027; reports already in progress keep their old deadlines.
Three clocks get faster
Two other timelines shrink along with the disclosure window:
- If a maintainer doesn’t respond to the first report, Anthropic will escalate to an external vulnerability coordinator after 15 days instead of 30.
- Full technical details come out 30 days after a patch is available, down from 45.
- The seven-day window for bugs under active exploitation stays as it is.
There’s also a catch-up step. Some high- and critical-severity bugs Anthropic reported more than 90 days ago are already past the current deadline and still not public. Their maintainers are being told now: 30 more days, then the reports go on Anthropic’s dashboard under the new policy from January 1.
The reasoning: what one model finds, another finds soon
The old norms were written when fixes moved at human speed and nobody expected two people to stumble on the same bug independently. Anthropic says neither assumption holds anymore. Models help write the fix, and a bug one model can find is increasingly likely to be found by another. The company says it’s seeing more rediscovery as capabilities improve. A 90-day window assumes a vulnerability stays private for months, and Anthropic no longer believes it does.
There are exceptions. Hard-to-deploy software, such as critical infrastructure, and flaws that need a change to a specification or RFC get a remediation timeline negotiated with the vendor. Findings from OSS Scanner that no human has reviewed aren’t covered by the new policy.
What maintainers get in return
Anthropic acknowledges that shorter deadlines put more pressure on maintainers. Each report will keep including actionable guidance and, where possible, a patch or mitigation. Before sending a large batch of findings to one project, Anthropic agrees on a pace with its maintainers first. Critical open-source projects can apply for OSS Scanner, a free service that regularly scans their code with Anthropic’s strongest models. Maintainers can also apply for six months of Claude Ultra through the Claude for Open Source Program, and security professionals for the Cyber Verification Program.
The clock is on the defenders now
On the same day, Google’s Threat Intelligence Group described the first attack framework that turns patches into exploits with no human involved. That’s exactly the situation Anthropic is reacting to. If you maintain an open-source project, you’ll have a month for a critical bug starting in January. The bet behind it: the bug won’t stay secret for long anyway, whether Anthropic publishes it or an attacker’s model finds it on its own. For small projects run by a few volunteers, that’s tight. The patch that comes with each report is what will make the deadline workable for them.