Google Project Zero and Anthropic are shortening their vulnerability disclosure windows as AI makes independent rediscovery more likely in their assessment. Software teams should measure how long it takes to protect users, including the time it takes them to get a fix into production.
Imagine that a researcher privately reports a serious flaw in the on-premises edition of your product. Engineering can produce a patch in two weeks. Your largest customers need a month to test and install it, and some still run a release line that your current build pipeline can barely reproduce. If the researcher will publish on day 30, the question for your CTO is simple. Which customers will actually be protected by then?
Hexkey’s ARTEX analysis looked at the speed of attacks against software already in service. The latest disclosure announcements raise a different question for the companies that build and maintain that software. How much time does your release process leave customers to act? Hexkey’s ARTEX analysis
What changed in November
On November 13, Google Project Zero replaced its “90 + 30” policy with a fixed report-to-disclosure clock. The earlier policy allowed up to 90 days for remediation, with a separate period for patch rollout. The new clock starts at 90 days, drops to 60 on February 2, 2027, and reaches 30 on January 3, 2028. The ordinary deadline runs from the report regardless of whether a fix ships. Project Zero retains exceptions, including separate handling for vulnerabilities under active exploitation. Project Zero’s disclosure announcement
On November 16, Anthropic announced that for reports sent from January 1, 2027, it will shorten its standard window from 90 days to 30 for high- and critical-severity issues and 45 for other issues. Its seven-day window for active exploitation stays in place. The change applies to findings that have received human review, and Anthropic says it will work with vendors on timelines for certain hard-to-remediate issues. Anthropic’s disclosure announcement
These are the reporting organizations’ own policies. Their shared rationale is worth taking seriously. Anthropic reports seeing more independent rediscovery. Project Zero says AI is changing how quickly attackers can learn from public fixes. In separate threat reporting, Google Threat Intelligence Group found an attacker-operated server with tooling set up to analyze software patches automatically. Data from victims also on the server suggested the operator was targeting French websites in specific sectors. That finding supports planning for exposure around patch release. The pace will vary with the flaw and who is looking for it. Anthropic’s announcement · Google Threat Intelligence Group’s observations
Put actual customer protection on the timeline
A software maker controls several dates that tend to get collapsed into “time to patch.” The first is when the report reaches someone who can assess it. The second is when a tested fix or mitigation reaches each affected release line. The third is when the fix takes effect where customers actually run the product. A cloud service owner can often see that last event in deployment telemetry. A vendor shipping customer-managed software may have to estimate it from support data, customer contacts, or opt-in update reporting. Give each date its own owner and record where you have direct evidence.
Then work backward from the date customers need protection. In the opening scenario, shipping on day 14 leaves 16 days for adoption under a 30-day deadline. For customers who need a month, look for a temporary control they can deploy sooner and validate against their setup. That might mean disabling an optional feature, restricting a management interface, or changing a supported configuration. Give customers the expected impact on normal use, a way to confirm the mitigation took effect, and a clear point at which to remove it after upgrading. NIST’s patch-management guide treats emergency mitigation and emergency patching as distinct operating scenarios, which is a useful distinction here. NIST’s enterprise patch management guide
Put the supported release lines into the exercise early. A fix in the current version may be straightforward. Backporting it can mean different dependencies, a different installer, and a different release owner. Find out which branch eats the available time before a live disclosure forces the issue. If a meaningful customer cohort depends on that branch, test the backport path or put an upgrade and mitigation plan in front of those customers while there is time to act.
Give customers a role in the exercise. Ask an account team to walk through what happens when an advisory reaches a customer with a weekend change freeze or a managed service provider in the middle. The advisory should make clear which editions are affected, who can make the change, what service loss to expect, and how to check the result. Identify customers who need direct outreach based on what they run and the likely impact. When you lack deployment telemetry, agree on how to confirm protection with those customers. Otherwise the team can hit its internal release date without knowing whether the people at greatest risk received the fix.
Buy back time where the exercise stalls
Take one plausible high-impact report and run a tabletop with security, release engineering, and customer support. Time the handoffs. Can somebody acknowledge the report, find the right code owner, identify affected versions, and get a candidate into a representative environment? Who can issue a customer advisory while release engineering is still working? Ask support to explain how a customer would tell whether its own installation is protected. Where the answer depends on somebody’s memory or weekend availability, you have found an investment to consider.
That investment may be modest. If a report sits in a general inbox, fix intake and escalation. If a release waits for someone to reconstruct an older build, fund reproducible release tooling. If customers take weeks to schedule upgrades, a safe, documented mitigation or a better update mechanism can save more exposure time than another tool that writes code.
There is a real tradeoff here. Project Zero says vendors warned that aggressive deadlines could divert effort from more severe issues. Keep the researcher’s deadline visible alongside your own assessment of customer impact, and give the release owner authority to resolve that conflict. For the next budget conversation, bring the slowest handoff from the tabletop and the customers it leaves exposed. That is the part of the disclosure window you can buy back. Project Zero’s discussion of vendor feedback
Sources and further reading
- Hexkey, What ARTEX changes for software you already run, November 20, 2026.
- Natalie Silvanovich, Google Project Zero, Policy and Disclosure: 2026 Edition, November 13, 2026.
- Anthropic Frontier Red Team, Updating our coordinated vulnerability disclosure policy timelines, November 16, 2026.
- Aurora Blum, Kelli Vanderlee, Jess Xia, and David Mainor, Google Threat Intelligence Group, GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold, November 16, 2026.
- Murugiah Souppaya and Karen Scarfone, NIST, Guide to Enterprise Patch Management Planning, April 2022.
Want a second look at your own applications?
We find and verify what’s exposed, fix what matters, and modernize what can’t be secured piece by piece.

