Anthropic vulnerability disclosure deadline drops to 30 days for serious bugs

Anthropic will publicly disclose high- and critical-severity vulnerabilities it reports to vendors and open source maintainers after 30 days, down from 90. Every other bug it reports gets 45 days. The new deadlines cover reports sent on or after January 1, 2027, and reports already open on that date keep the deadlines they started with.

Anthropic vulnerability disclosure

The company says it no longer expects a reported vulnerability to stay private for months, because a bug one model finds is increasingly likely to be found by another. For maintainers, the change cuts the time to triage, fix, and ship a patch for a serious flaw by two thirds. For people running that software, details of a flaw can reach the public sooner, on the dashboard where Anthropic publishes its reports.

Two more clocks got shorter

If a maintainer does not answer the first report, Anthropic will aim to escalate to an external vulnerability coordinator, a third party that brokers contact between the finder and the vendor, after 15 days, down from 30. Once a patch is available, Anthropic will wait 30 days before releasing full technical details, down from 45.

Put side by side, the old and new numbers are 90 and 30 days for serious bugs, 90 and 45 for everything else, 30 and 15 days before escalation, and 45 and 30 days between a patch and full details. The seven-day window for vulnerabilities under active exploitation stays where it was. Anthropic says the rest of its coordinated vulnerability disclosure (CVD) policy, the rules it follows for warning a vendor privately before going public, is unchanged.

Reports already past due

Anthropic is holding high- and critical-severity reports it filed more than 90 days ago that have passed the deadline in its current policy and are still undisclosed. It is notifying those maintainers that they have 30 more days, after which it will publish the reports on its dashboard, in line with the updated policy.

Those flaws have already outlasted a deadline three times as long as the one Anthropic is about to adopt.

Why Anthropic moved

Anthropic’s policy, like many disclosure norms, was written before AI could find vulnerabilities. “These policies assumed that fixes moved at human speed and that independent rediscovery of the same bug was unlikely. Neither assumption holds today,” the company wrote.

Models help write fixes, and Anthropic says it is seeing more rediscovery of the same vulnerabilities as model capabilities grow.

Where 30 days won’t hold

Anthropic expects some fixes to run long. Two groups stand out: flaws in critical infrastructure and other software that is hard to deploy, and flaws that can only be closed by changing a specification or an RFC, the documents that define internet protocols. For those, Anthropic will work out a remediation timeline with the vendor.

The new policy also does not apply to AI-discovered findings that have not had human review, including findings from OSS Scanner, Anthropic’s free code scanning service. Anthropic did not say which deadline those findings follow.

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, so the maintainer can confirm the finding and fix it. Before sending a large batch of findings to one project, Anthropic will keep agreeing on a workable rate with its maintainers.

OSS Scanner is open by application to maintainers of critical open source projects. It runs Anthropic’s strongest models over their code on a recurring basis, for free, and sends the findings straight to them. Eligible maintainers and contributors can apply to the Claude for Open Source Program for six months of Claude Ultra at no cost. Security professionals, open source maintainers included, can apply to the Cyber Verification Program, which adjusts Claude’s safeguards to allow legitimate defensive work.

Don't miss