AI

Anthropic 30-day disclosure deadline: how to get patch-ready

Key takeaways

  • High- and critical-severity bugs reported by Anthropic go public after 30 days, down from 90, for reports sent on or after 1 January 2027
  • Full technical details now follow 30 days after a patch, not 45, and escalation to a coordinator comes after 15 days of silence
  • Overdue serious reports get 30 more days, then appear on Anthropic's dashboard
  • Maintainers and IT teams should set up a monitored security contact and a 30-day triage routine now

After reading this, a maintainer or IT lead will be able to set up a workflow that can triage, fix and ship a patch for a serious vulnerability inside 30 days, and to track what that means for the software they run. Difficulty: intermediate. Time to set up: roughly half a day for a small project, a few days for an organisation with several teams.

Anthropic has cut the public disclosure deadline for high- and critical-severity vulnerabilities it reports to vendors and open source maintainers from 90 days to 30. Every other bug gets 45 days. The change covers reports sent on or after 1 January 2027, and reports already open on that date keep their existing deadlines.

What you need before starting

  • A named security contact for each project or product, with a mailbox that someone reads daily.
  • A private way to receive reports and share patches (a private issue tracker, a security advisory feature on your code host, or an encrypted mailbox).
  • A severity rating habit, so "high" and "critical" can be separated from everything else within a day or two.
  • A person with authority to approve and ship an emergency release.
  • Cost: nothing for the process itself. Maintainers of critical open source projects can apply for Anthropic's OSS Scanner, which runs its strongest models over their code on a recurring basis, for free. Eligible maintainers and contributors can also apply to the Claude for Open Source Program for six months of Claude Ultra at no cost.

Step-by-step process

Step 1: Know the new clocks (10 minutes)

Write the numbers on one page. For Anthropic's reports: 30 days to disclosure for high and critical severity, 45 days for everything else, escalation to an external coordinator after 15 days without a maintainer response (down from 30), and full technical details released 30 days after a patch is available (down from 45). The seven-day window for vulnerabilities under active exploitation is unchanged. Anthropic says the rest of its coordinated vulnerability disclosure policy, the rules for warning a vendor privately before going public, is unchanged.

Step 2: Make sure the first message reaches a human (30 minutes)

Anthropic aims to escalate to a coordinator after 15 days without a maintainer response to its first report. A security contact that bounces, or a mailbox nobody checks, now risks a third party being brought in within a fortnight. Publish a security contact file or policy page, test it by sending a message from an outside address, and set an auto-acknowledgement that confirms receipt.

Step 3: Build a 30-day triage plan (1 to 2 hours)

Split the 30 days into stages. A workable pattern: acknowledge within days, confirm and rate severity in the first week, develop and review the fix in the middle, and keep the final days for testing and release. Anthropic says each report will keep including actionable guidance and, where possible, a patch or mitigation, so the confirmation step should be faster than starting from a bare description.

Step 4: Agree a workable rate for batches (30 minutes)

If several findings may arrive together, tell Anthropic what volume you can absorb. The company says that before sending a large batch of findings to one project, it will keep agreeing on a workable rate with its maintainers. Use that.

Step 5: Flag the exceptions early (as needed)

Anthropic expects some fixes to run long: flaws in critical infrastructure and other hard-to-deploy software, and flaws that can only be closed by changing a specification or an RFC, the documents that define internet protocols. For those, it will work out a remediation timeline with the vendor. If your software fits, say so in your first reply, with reasons, rather than near the end of the 30 days.

Step 6: Prepare downstream, if you only run the software (1 hour)

Details of a flaw can now reach the public sooner, on the dashboard where Anthropic publishes its reports. The NCSC's 10 Steps to Cyber Security is a sensible baseline, and the zero-day checklist for network devices shows what a same-day response looks like in practice.

Common mistakes to avoid

  • Treating the old 90 days as the plan. Anthropic is also holding overdue reports to a deadline. Serious reports it filed more than 90 days ago that are past due and still undisclosed get 30 more days, then publication on its dashboard. Maintainers holding such a report are already on the clock.
  • Assuming silence buys time. Escalation to a coordinator after 15 days is a push to engage, not a penalty to wait out.
  • Assuming every AI-found bug follows this policy. Findings that have not had human review, including those from OSS Scanner, are outside it, and Anthropic did not say which deadline those follow.
  • Confusing severity with exploitation. A bug under active exploitation keeps the seven-day window whatever its rating.
  • Waiting for a patch to start communicating. The 30 days after a patch before full technical details are released comes after a patch is available, so announce the fix clearly.

Tips that make a difference

  • Anthropic's reasoning is that a bug one model finds is increasingly likely to be found by another. Treat a private report as a head start that may be short, and do not assume nobody else has seen the flaw.
  • Ask for the patch or mitigation if one is not attached. Anthropic says it will include one where possible.
  • Security professionals, open source maintainers included, can apply to the Cyber Verification Program, which adjusts Claude's safeguards to allow legitimate defensive work. That is aimed at legitimate defensive work.
  • Google's Project Zero, which now publishes details 90 days after reporting a bug whether or not it is fixed, will cut that to 60 days on 2 February 2027 and 30 days on 3 January 2028. Plan for shorter windows from more than one vendor.

What this means for UK organisations

The NCSC publishes incident management guidance that is a useful template for a rapid patch-and-communicate routine. Anthropic's policy itself does not create any UK obligation; the pressure comes from public details arriving sooner.

Next steps

  • Run a table-top exercise using a fictional critical report on day 1, day 15 and day 30.
  • Compare how fast your team patched the last serious vulnerability against a 30-day target, and record the gap.
  • Read how GPT-6 Cyber and similar tools are changing discovery volumes, since higher volumes make triage capacity the bottleneck.
  • Check which of your own dependencies are actively exploited using the CISA KEV catalogue.

Anthropic has moved from 90 days to 30 for serious bugs from 1 January 2027, and shortened its escalation and post-patch clocks too. Maintainers get actionable guidance and, where possible, a patch with each report. The practical work is mundane: a monitored contact, a staged 30-day plan and an early conversation when a fix will genuinely take longer.

Summary

You now have a named security contact, a staged 30-day triage plan, a route for flagging exceptions and a downstream update routine, which together cover the new deadlines before they take effect on 1 January 2027.

Frequently asked questions

When does Anthropic's new disclosure deadline apply?

It applies to vulnerability reports sent on or after 1 January 2027. Reports already open on that date keep the deadlines they started with.

How long will Anthropic wait before disclosing a serious bug?

Thirty days for high- and critical-severity vulnerabilities, down from 90. Every other bug gets 45 days. Flaws under active exploitation keep the seven-day window.

What happens if a maintainer does not respond to Anthropic?

Anthropic will aim to escalate to an external vulnerability coordinator after 15 days without an answer, down from 30. A coordinator is a third party that brokers contact between finder and vendor.

Does the 30-day deadline cover OSS Scanner findings?

No. The policy excludes AI-discovered findings that have not had human review, including those from OSS Scanner. Anthropic did not say which deadline those findings follow.

More from Future Technology HQ