Skip to content
RESEARCH INDEX BREACHROAD / INTELLIGENCE NOTE

Google Project Zero Moves to 30-Day Disclosure

Project Zero is replacing 90+30 with Full 90, then 60 and 30 days. We explain the schedule, patch gap, AI rationale and vendor readiness.

PUBLIC RESEARCH
AUTHOR
/ CEO Breachroad · OSCP · PNPT
PUBLISHED
17 November 2026
READING TIME
21 min read
TOPIC
Vulnerabilities and CVEs
Google Project Zero Moves to 30-Day Disclosure

Google Project Zero is changing its coordinated vulnerability-disclosure policy. The team has replaced “90+30” with a fixed Full 90 deadline, plans to move to 60 days on 2 February 2027, and to 30 days on 3 January 2028. Once the applicable deadline expires, technical details will be published whether or not the vendor has released a fix.

The change was announced on 13 November and became a major product-security topic on 16–17 November. The official Project Zero disclosure policy does not give every vendor 30 days immediately. The transition is gradual, and each report is governed by the policy in force on its reporting date.

That distinction is essential for accurate reporting and SEO. “Google cuts disclosure to 30 days” describes the direction, but a useful analysis needs the exact dates, exceptions and patch-gap consequences.

The Full X schedule

The new clock is simple:

Vulnerability report datePolicyPlanned disclosure
13 November 2026–1 February 2027Full 9090 days after report
2 February 2027–2 January 2028Full 6060 days after report
from 3 January 2028Full 3030 days after report

Project Zero will disclose details after X days without making the date dependent on the patch. If a vendor fixes a bug on day 47, disclosure still occurs on day 90 under the current policy. If no patch exists on day 90, the details are also released.

The team will continue tracking remediation and communicating whether users are protected. Fix status will no longer be a decision point for the clock. That simplification makes the policy easier to automate for both Project Zero and vendors.

What disappeared from 90+30

Since 2021, Project Zero gave vendors 90 days to remediate and an additional 30-day “soak time” when a fix arrived before the deadline. The extra period was intended for patch distribution and adoption. Disclosure could therefore occur up to 120 days after the initial report.

Full X removes most special cases, including the grace period and variant policy. Project Zero argues that rapid patch analysis makes post-release secrecy increasingly artificial and can give users a false sense of safety.

In the official policy rationale, Natalie Silvanovich writes that AI has reduced the time required to reverse engineer and exploit a vulnerability from its patch. The team considers “zero patch-to-exploitation time” a reasonable, although imperfect, model of attacker behaviour. A release process should therefore assume that analysis begins immediately when the patch appears.

Exceptions still exist

Full X is not a mechanical rule without exceptions. The policy identifies:

  • vulnerabilities actively exploited against real users, which follow Google’s separate application-security policy and generally a seven-day deadline;
  • public vulnerabilities whose details have already been released by another party;
  • earlier disclosure requested by the vendor;
  • rare exceptions where disclosure would risk widespread harm and delay would materially help prevent it.

Google says exceptions are rare and publicly recorded. A complex hardware product should not assume that difficulty alone guarantees more time. Vendors need non-patch protections: disabling a function, changing service configuration, applying a server-side rule or another control that protects users before a full device update arrives.

Why Project Zero points to AI

Project Zero describes three pressures. First, LLMs increase vulnerability-discovery and exploitation speed, along with the chance that multiple parties—including malicious ones—find the same bug. Second, differential patch analysis can reveal the affected area and defect mechanics almost immediately. Third, AI has increased the number of vulnerabilities the team itself reports.

This is not evidence that all exploits are produced autonomously. It is a process decision based on risk: organisations should not rely on a patch remaining unanalyzed for weeks. The GTIG report on autonomous exploit_agent operations provides additional operational context.

Patch gap matters more than the commit date

Patch gap is the time between a fix being available and a real user becoming protected. In an upstream-to-downstream chain it may involve a library maintainer, system vendor, distribution, integrator, device operator and end administrator.

For example, an open-source project fixes a driver on day 20. A chipset vendor integrates it on day 30, a phone manufacturer builds firmware on day 45, a carrier approves it on day 60 and a user installs it on day 75. “Fixed in 20 days” does not describe the real risk window.

Since 2025, Project Zero has operated Reporting Transparency. Within roughly a week of a report, it publishes the report identifier, vendor or project, product, report date and deadline—without technical detail, proof-of-concept code or information that materially helps discovery. The early signal is intended to warn downstream dependants before full disclosure.

What 30 days means for a software vendor

Thirty days leaves no room for a purely sequential process: triage, owner discovery, analysis, implementation, testing, documentation, packaging, deployment and only then communication. Some work must happen in parallel.

Before a report arrives, a vendor needs:

  • a functioning security contact and CVD policy;
  • on-call coverage and backup owners for components;
  • a map of products and downstream versions;
  • reproducible builds and a fast security pipeline;
  • regression tests for critical paths;
  • an out-of-band update mechanism;
  • signed artefacts and automated rollback;
  • prepared advisory, CVE and customer-communication templates;
  • metrics showing whether a patch reached users;
  • an alternative mitigation when a full patch requires more time.

Without those foundations, the first days are lost to organisation rather than remediation. DevSecOps security in CI/CD must include an emergency mode, not just a monthly release train.

Quality triage, not a queue ordered by noise

Project Zero notes that it reports vulnerabilities with substantial severity and demonstrated impact. Its reports aim for high reproducibility, readability and reachability analysis. Full X is designed around that quality of input.

A vendor receiving low-quality automated submissions should not treat every report identically. Fast triage asks whether a report identifies the version, prerequisites, evidence, impact, reproducibility and a safe contact. AI can increase both genuine discoveries and false positives.

The “AI slop” problem must not become an excuse to ignore a rigorous report. Technical quality—not reputation or an emotional response to the reporter—should determine treatment.

How to remediate faster without lowering quality

Speed does not mean skipping tests. It means investing in advance and limiting the change. A security patch should be small, tested, reviewed and deployed through a canary. A broad refactor under deadline pressure raises regression risk.

A practical workflow is to:

  1. acknowledge the report and assign an owner within hours;
  2. reproduce the issue in an isolated environment;
  3. map every affected product and branch;
  4. create a test that first demonstrates the defect;
  5. implement the smallest effective change;
  6. run functional and security tests;
  7. prepare advisory and mitigation in parallel;
  8. deploy a canary and monitor telemetry;
  9. sign and publish artefacts for every supported release;
  10. measure installation rather than download alone.

When a patch cannot be ready, restrict the feature, change service configuration or apply a compensating control. Project Zero explicitly recognises non-patch strategies when they genuinely protect users.

What enterprise customers should change

The policy governs researcher–vendor interaction, but enterprise customers will see less time between an early signal and technical disclosure. Vulnerability-management teams should monitor Reporting Transparency, vendor channels, CISA KEV, national CSIRTs and their own SBOMs.

They need an emergency-change path. Not every patch belongs there, but a critical flaw in an Internet-facing system cannot wait for a monthly change board. The process must identify who accepts risk, the minimum test, canary, rollback and independent monitoring.

Useful metrics include:

  • time to identify affected assets;
  • time to choose patch or mitigation;
  • time to first and complete deployment;
  • number of exceptions beyond the deadline;
  • mismatch between production and DR;
  • percentage of assets with a credible owner;
  • time to retest.

Our guide to business-driven vulnerability prioritisation develops the model.

Hardware, medical and critical systems

Products with long certification cycles—firmware, medical devices, vehicles and industrial systems—are the hardest cases. A complete update in 30 days may be unrealistic, but leaving users unprotected is not the only alternative.

Vendors should design feature kill switches, server-side rules, segmentation, remote configuration, monitoring and secure modular updates in advance. Customers need compensating-control guidance that does not invalidate device certification.

Project Zero selected a gradual transition because immediate Full 30 could encourage some vendors to give up or rush unsafe fixes. The schedule is intended to be challenging while allowing time to rebuild remediation infrastructure.

CVD and the researcher relationship

A shorter deadline makes coordinated vulnerability disclosure more important. An automated acknowledgement is not enough. Researchers need a person, case number, secure channel and status updates that do not expose sensitive detail.

A dispute over severity should not delay reproduction. CVSS describes part of the risk but does not settle reachability, exposure or impact in a real product. Project Zero does not vary the deadline by severity because impact does not change the chance of independent discovery.

Organisations should publish a reporting policy, safe harbour, encryption key and clear scope. That reduces the chance that a valuable disclosure disappears into a generic support form.

Why full disclosure can help defenders

Publishing detail after the deadline increases risk to unpatched systems, but it also lets defenders independently validate exposure, build detection and assess mitigation. Without disclosure, customers depend entirely on the vendor’s wording, which may omit conditions or impact.

Responsible disclosure does not need a turnkey destructive exploit. It can document root cause, versions, prerequisites, impact and a safe method for confirming the fix. The goal is to balance vendor pressure with user protection.

Source facts and Breachroad conclusions

Project Zero confirms Full 90 from 13 November 2026, Full 60 from 2 February 2027 and Full 30 from 3 January 2028. It confirms removal of the grace period, retention of narrow exceptions and continuation of Reporting Transparency. The team cites AI, patch-to-exploitation speed and rising report volume as reasons.

The SLA, emergency-change, metric, canary, CVD and regulated-product recommendations are Breachroad analysis. Organisations can validate readiness through an IT security audit and web and API penetration testing.

The key takeaway

Full 30 does not begin today, but preparation does. A vendor that starts building a security contact, hotfix pipeline and downstream map in 2028 will not meet a thirty-day deadline without quality risk.

The strongest metric is not the ticket-close date. It is the moment the end user is protected. Project Zero’s new policy pushes attention to that final mile: from corrected source code to an actually updated product.

SHARE / COPY