Skip to main content
Promotional banner ad for the Penetration Testing Report Kit
Anthropic's 30-Day Disclosure Clock: A Field GuideAdversarial Security
4 min readFor AI Governance Leaders

Anthropic's 30-Day Disclosure Clock: A Field Guide

Scope

This guide explains the changes in Anthropic's Responsible Disclosure policy for flaws identified by its Claude models. It compares these changes to Google's Project Zero and outlines adjustments your team should make in vendor management, risk tiering, and incident workflows.

This policy doesn't cover bugs reported by external researchers to Anthropic. It applies to vulnerabilities Anthropic reports to vendors and maintainers, excluding AI-discovered findings without human review, like output from its OSS Scanner.

Applicability date: vulnerabilities reported on or after January 1, 2027. Reports already in progress by then will keep their existing deadlines.

Key Concepts and Definitions

Responsible Disclosure. Privately reporting a flaw to the maintainer and withholding public details for a set time to allow for a fix.

Disclosure deadline. The days from report to publication, regardless of patch status, unless specified otherwise.

Patch-to-detail gap. The time between a patch becoming available and the publication of full technical details.

External vulnerability coordinator. A third party Anthropic contacts if a maintainer doesn't respond.

Independent rediscovery. The likelihood that another party, human or model, finds the same bug. Anthropic notes that traditional assumptions about fix speed and rediscovery don't hold today.

Requirements Breakdown

These are the new parameters your vendors and open-source dependencies face when Claude-discovered bugs are reported.

Parameter Previous New (reports on/after Jan 1, 2027)
High/critical severity deadline 90 days 30 days
All other issues Not specified here 45 days
Escalation to external coordinator After 30 days without maintainer response Aim to escalate after 15 days
Wait after patch before full technical details 45 days 30 days
Active exploitation 7 days 7 days (unchanged)

Additional terms:

  • Anthropic will collaborate with vendors on a remediation timeline for hard-to-deploy software, such as critical infrastructure, and for flaws needing specification or RFC changes.
  • Anthropic will provide actionable guidance with each report, plus a patch or mitigation where possible.
  • Maintainers get 30 more days for high- and critical-severity bugs that have passed the current 90-day deadline before publishing them on Anthropic's disclosure dashboard.

Benchmark: Project Zero

Project Zero currently publishes details 90 days after reporting a bug, fixed or not. This schedule tightens to 60 days on February 2, 2027, and 30 days on January 3, 2028.

Key takeaways:

  1. Anthropic's serious-bug deadline is ahead of the benchmark. Its 30-day deadline applies a year earlier than Project Zero's.
  2. Different philosophies on severity. Anthropic varies windows by severity, while Project Zero argues severity doesn't affect how likely attackers are to find a bug. If you adopt a tiered internal SLA, you're aligning with Anthropic's model; document your reasons.

Implementation Guidance

  1. Rebase your remediation SLAs. If your internal patch SLA for high-severity findings in third-party software is 60 or 90 days, it no longer aligns with major disclosure sources. Set a target to ship or mitigate within 30 days, or document a compensating control path.

  2. Shorten maintainer response expectations. A 15-day escalation trigger means silence gets noticed early. Ensure someone monitors your security inbox and can acknowledge within that window.

  3. Update vendor contracts. Review security addenda for notification and remediation clauses based on 90-day norms. Add language for vulnerability intake from AI-assisted discovery, acknowledgment timelines, and coordination responsibilities when a deadline is at risk. Ask suppliers how they'll handle the 30-day patch-to-detail gap, as your exposure window after a patch is shorter too.

  4. Pre-plan for exceptions. The hard-to-deploy carve-out is negotiated, not automatic. If you operate critical infrastructure or software tied to specification changes, prepare evidence of why standard timelines don't fit and propose a remediation schedule.

  5. Map to your risk framework. Add accelerated disclosure to your risk register under third-party and AI Supply Chain Compromise scenarios. Align treatment with ISO/IEC 23894 risk identification and the NIST AI RMF Manage function. Record deadline changes as a trigger for control review and log Validation Evidence for emergency patches that skipped normal change windows.

  6. Rehearse the 7-day case. The active-exploitation window didn't change, but if more bugs are rediscovered quickly, more may fall into this category. Test your emergency patch path end to end.

  7. Legal review. Shorter public-disclosure deadlines can affect notification duties, customer commitments, and regulatory reporting timelines. Have counsel check whether contractual or statutory clocks in your sector now start closer to, or before, a vendor's patch. Treat this as an issue to scope with your own counsel.

Common Pitfalls

  • Assuming the policy covers everything. It governs what Anthropic reports to others, not what researchers report to Anthropic. Don't cite it as Anthropic's inbound intake policy.
  • Reading 30 days as a patch guarantee. It's a disclosure deadline. Details can be published even if your fix isn't ready, except where a negotiated timeline applies.
  • Ignoring the non-severity tier. Lower-severity issues still carry a 45-day clock.
  • Treating the date as retroactive. Reports in progress on January 1, 2027, keep their existing deadlines; plan the cutover by report date, not calendar alone.
  • Over-indexing on one source. Anthropic and Project Zero differ on severity logic and dates. Build policy that survives either approach.
  • Skipping the human-review boundary. Unreviewed AI findings fall outside this policy, so they follow different handling.

Quick Reference

Question Answer
Effective date Reports on or after January 1, 2027
High/critical deadline 30 days
Other issues 45 days
Escalation trigger 15 days without maintainer response (aim)
Patch-to-detail wait 30 days
Active exploitation 7 days, unchanged
In-progress reports Keep existing deadlines
Out of scope External reports to Anthropic; unreviewed AI findings
Project Zero today 90 days, regardless of fix
Project Zero next 60 days (Feb 2, 2027); 30 days (Jan 3, 2028)
Exception path Negotiated timeline for hard-to-deploy or spec-change fixes

Bookmark this, then check your vendor SLAs against the table. If your slowest high-severity path exceeds 30 days, that's your first gap to close.

Promotional banner highlighting failures found in PCI audits and how to spot the gaps

You Might Also Like