# Google's Project Zero cuts the deadline to patch vulnerabilities to 90 days

> Since November 13, Google has published the details of a flaw 90 days after the report, patch or no patch, and will cut that to 60 in 2027 and 30 in 2028.
> By José Maldonado · Cybersecurity · 2026-11-19

A flaw reported in January and published in April hands a three-month head start to the attacker who finds it on their own. Project Zero, Google's research team, has narrowed that margin: the details are published a fixed number of days after the report, whether or not there is a patch. It is called Full X and the team itself explained it in [the post announcing the policy](https://projectzero.google/2026/11/disclosure-policy.html), dated November 13, the day it took effect. Today it is 90 days; in 2027 it will be 60 and in 2028, 30.

The previous policy had stood for five years without substantive changes. The team gave the vendor 90 days to fix the flaw, added 30 days of margin for the patch to reach devices and allowed a two-week extension if the provider requested one. In the worst case, 120 days passed between the report and publication, a buffer designed so the update would spread before anyone knew what to look for.

That buffer disappears because it no longer covers anything. Language models speed up flaw discovery and make it possible to reconstruct the vulnerability from the patch in minutes; when an attacker can read the fix, keeping the analysis quiet for a month only creates a false sense of secrecy. Project Zero has just written that silence comes too late.

## What exactly changes and when

Full X means what the name says: publication X days after the report, unconditionally. Full 90 already covers any flaw notified since November 13. Publishing with the flaw still open is the normal scenario. A vendor can end up with 33 zero-days in its product and the cloud down, as [N-able did with N-central](https://observatorioblockchain.com/en/cybersecurity/able2shell-n-central-33-cero-dias-nube-offline/), and the calendar does not move because of that.

Stage

Takes effect

Period between report and publication

Full 90

November 13, 2026

90 days

Full 60

February 2, 2027

60 days

Full 30

January 3, 2028

30 days

More pieces fall than just the deadline from the previous policy: the 30-day grace period, the two-week extensions and the special rules for variants of an already known flaw. [heise online goes through them one by one](https://www.heise.de/en/news/Google-s-Project-Zero-shortens-vulnerability-disclosure-deadlines-11528434.html).

Three exceptions remain and none works as an extension. A flaw that is already being exploited moves to Google's general policy, which gives a zero-day seven days to publish mitigations under [Google's application security policy](https://about.google/company-info/appsecurity/). A flaw that others have made public is counted whenever the team wants. And the vendor keeps one route: asking for disclosure to be brought forward.

The least discussed change is the one on severity. Project Zero no longer uses a flaw's score to decide when to publish it, because severity measures potential damage, not the likelihood that someone will find it. A low-severity flaw is still available to anyone who looks for it. That makes vendors uncomfortable: they can no longer defend a delay by saying the flaw was minor.

## Why AI changes the calculation

Project Zero finds several vulnerabilities a week where there used to be a few a year. The team documents two distinct effects and they should not be conflated: the first is that volume, and the second is the speed of exploitation, which affects every market participant, not just Google.

[Natalie Silvanovich, a researcher on the team, sums it up in CSO Online](https://www.csoonline.com/article/4249915/google-and-anthropic-push-for-faster-patches-as-ai-collapses-bug-discovery-timelines.html): models have increased the speed at which an attacker finds and exploits flaws, and they multiply the risk that several parties, including malicious ones, discover the same bug. A flaw reported today is a much greater immediate risk than one reported last year.

The scale can already be measured outside Google. [The 948 flaws OpenAI found in the Linux kernel with AI agents](https://observatorioblockchain.com/en/ai/openai-detecta-948-fallos-linux-tomar-control/) came with proposed patches and human review behind them, and that is the volume now coming through a vendor's door.

If reports multiply and deadlines shrink, the limit stops being the willingness to patch and becomes the capacity to triage. A two-person team that receives three times as many alerts and has half the time does not fix faster: it prioritizes worse.

## The deployment margin was a promise nobody kept

The strongest argument for the change is empirical, not technological. The 30-day grace period existed so the patch would reach systems before the flaw became known, and that premise was never fully met. Most organizations take weeks or months to apply an update, and an extra month of silence does not make up for that delay.

The pattern repeats in every update window, as in [Windows 11 26H2, which marks a week with flaws and half-finished patches](https://observatorioblockchain.com/en/technology/windows-11-26h2-fallos-una-semana/): publishing earlier is not the problem, installing later is.

Project Zero puts it in a sentence worth reading twice: if an attacker can learn the details of the flaw almost instantly from the patch, keeping the analysis private offers little real protection. The effort should go into closing the gap between publication of the fix and its installation. That is the patch gap, and it is measured in days of exposure.

There is a detail the team itself acknowledges: disclosure is an incentive, not a guarantee. A vendor unable to fix within 30 days will not gain that ability because it is asked to, and that is why the calendar is staggered. The first step has a date: February 2, 2027, at 60 days.

## Who loses with the new calendar

The most cited criticism comes from Marcus Hutchins, the researcher who stopped WannaCry in 2017. His objection, reported by [IT Pro](https://www.itpro.com/security/companies-will-be-given-30-day-patch-deadlines-as-researchers-tighten-bug-disclosure-windows), does not dispute the diagnosis but the burden: maintainers of open projects are already overwhelmed by the increase in reports and a shorter deadline could burn them out.

Project Zero's response is that volunteer projects are not the problem, because almost all flaws reported this year were fixed in under 30 days, with a single exception that took 31. It is the team's own figure, not an external auditor's, and it leaves little room for the objection.

The other affected group is vendors with long cycles. A product with extended support or a certified medical device does not change a version in 90 days even if it wants to. The case of [MPTCP, with CVE-2026-98432 patched in the stable branches and unpatched in RHEL](https://observatorioblockchain.com/en/cybersecurity/cve-2026-98432-mptcp-linux-sin-parche/) shows that the same fix arrives at different speeds depending on who packages it.

For those cases, intermediate mitigations remain, and the team accepts them as legitimate: disabling a feature, closing a port or restricting access count as a valid response at first. For a company that buys software and does not make it, the calendar affects what is published, not what is fixed.

## What can be done starting Monday

The first task is not technical. It is worth knowing which vendors have a written patching commitment, with deadlines, and which respond with an "as soon as possible" that means nothing. With a public calendar of 90, 60 and 30 days, that question can go into a contract.

The second is to accept that exposure is measured not in deadline days but in accessible services. A flaw in an internal system that no one can reach from outside does not have the same priority as one in a payment gateway, and that distinction existed before this policy.

The third is hygiene. If you manage Windows devices and servers that no one inventoried, no deadline helps you, neither 120 days nor 30. Google's policy does not fix that hole: it makes it visible at the worst moment.

## The calendar as security policy

What Project Zero has put on the table is not just a calendar, it is a thesis: disclosing a flaw is an economic decision, and whoever delays it is subsidizing the attacker with time. In 2028, at 30 days, a slow vendor's margin will be the same as a fast one's.

For those who administer systems, the message fits in one line: other people's deadlines are shrinking and your own are not. Preparing the inventory, the patching channel and an intermediate response that is not waiting is the only thing that depends on you when someone else sets the date.
