Anthropic Cuts Disclosure to 30 Days: What Maintainers Do

TL;DR
Anthropic is cutting its vulnerability disclosure window from 90 days to 30 for high and critical bugs, effective January 1, 2027. What changes, what does not, and a Monday checklist for maintainers.
Last updated: November 17, 2026
Anthropic's Frontier Red Team announced on November 16 that it is cutting its coordinated vulnerability disclosure window from 90 days to 30 for high and critical issues, and to 45 days for everything else. If you maintain an open-source project that Claude-class models might scan, this is the date to put on your calendar: January 1, 2027.
What the primary source says#
From Anthropic's post:
- Window: 30 days for high and critical severity, 45 days for all other issues, down from 90.
- Effective date: reports sent to vendors and maintainers on or after January 1, 2027. Reports already in progress keep their existing deadlines.
- Escalation: Anthropic aims to escalate to an external vulnerability coordinator after 15 days without a maintainer response, down from 30.
- After a patch: full technical details are shared 30 days after a fix ships, down from 45.
- Overdue backlog: high and critical reports more than 90 days old that are still undisclosed get 30 more days, then go on Anthropic's dashboard.
- Unchanged: the seven-day window for actively exploited bugs.
- Exceptions: hard-to-deploy software such as critical infrastructure, and flaws that need a spec or RFC change, get a remediation timeline agreed with the vendor.
The stated reason is that models have changed two assumptions behind the old norm. Anthropic says fixes no longer move only at human speed, and that one model's discoveries are increasingly likely to be found independently by another. In its words, the 90-day window was built for an era when a bug could stay private for months. Both are Anthropic's own claims and I could not verify the rediscovery rate independently.
The angle: who carries the cost#
The loser is the underfunded maintainer. A 30-day clock on a critical bug is tight for a volunteer project with no release automation, and the escalation trigger at 15 days of silence means a missed email can start the process of outside coordination.
The winner is whoever ships fast. Anthropic says each report will include actionable guidance and, where possible, a patch or mitigation. It also says it will agree on a sustainable pace before sending a large volume of findings to one project. Those commitments matter more than the number, so hold the company to them.
The scope matters too. The policy does not apply to AI findings that have not had human review, such as output from the OSS Scanner, a no-cost service that periodically scans critical open-source projects with Anthropic's strongest models. For context on where these findings come from, see Claude Security continuous scanning and what Claude Mythos 5 is and who can access it.
What a maintainer does Monday#
- Publish a security contact. Add a
SECURITY.mdwith a monitored address and a private reporting route. Anthropic's 15-day escalation clock starts from silence. - Decide who triages. One named person, plus a backup, for the 30-day window.
- Make a release cheap. If cutting a patch release takes a day of manual work, fix that before January.
- Check your own exposure. Agents leak secrets too; see GitGuardian's count of Claude Code settings leaks.
- Apply for support. Anthropic lists a Claude for Open Source Program (six months of Claude Ultra for eligible maintainers) and a Cyber Verification Program for security professionals. Eligibility is Anthropic's call; check the post for details.
If you consume open-source dependencies rather than maintain them, the practical shift is shorter gaps between a private report and public details. Review how quickly your pipeline picks up patched versions, and read the agent security checklist for the tool side.
What people are actually saying#
I did not find a Hacker News or Reddit thread about this specific announcement within a day of it going out, so there is no community reaction to quote yet. The nearest signal is older. A VulnCheck write-up on Anthropic's earlier disclosure ledger argued the previous policy was hard to read as a single clock, with a 90-day window, a 45-day post-patch wait, extensions and exceptions all in play. I only saw that critique in a search summary, not the full page, so treat it as a pointer. The counter-case to a shorter window is the one above: a short deadline favours well-resourced vendors over small maintainers. The case for it is that fast rediscovery makes long embargoes less protective than they look.
FAQ#
What is Anthropic's new vulnerability disclosure window?#
Thirty days for high and critical severity issues and 45 days for everything else, down from 90. It applies to vulnerabilities reported to vendors and maintainers on or after January 1, 2027.
Does the 30-day window apply to actively exploited bugs?#
The seven-day window for actively exploited vulnerabilities is unchanged, per Anthropic.
Do the new timelines cover every AI-found bug?#
No. Anthropic says the policy does not apply to AI-discovered findings that have not had human review, such as those from its OSS Scanner.
Continue Reading#
- Claude Security continuous scanning - fix tracking, API and MCP.
- Claude Mythos 5 explained - why access is gated.
- Claude Code secret leaks - the GitGuardian numbers.
- Agent security checklist - before you connect tools.
- Anthropic OSS Scanner enrollment - the free scanning service the policy excludes.
Sources#
Get the next deep dive like this in your inbox
One email a week on Security and the rest of the AI dev stack. Free.
Read next on Claude Code
Claude Security Continuous Scanning: Fix Tracking, API and MCP
Claude Security now rescans repos as commits land, marks findings fixed when the fix ships, and adds an API, MCP connector and webhooks. Setup, limits and cost.
8 min readClaude Mythos 5 Explained: What It Is, Who Can Access It, and Why It's Gated
Claude Mythos 5 explained - how it differs from Fable 5, who gets unrestricted access through Project Glasswing, and what developers should do right now.
7 min readClaude Code GitHub Secret Leaks: GitGuardian's 31,088
GitGuardian counts 31,088 secrets in Claude Code settings files on public GitHub, 5,343 valid, plus copies baked into Docker Hub images. What its numbers add to Apiiro's, and the Monday checklist.
7 min readTechnical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.








