Skip to content
HomeBriefFor youTrackersTopicsCompaniesGlossaryNewsletterSecurityHealth & medicineFinance & marketsRoboticsModels & researchPolicy & regulationEnterprise & businessDeveloper tools
Security · Nov 16, 2026 · 3 min read

Project Zero's 'Full X' policy: 90 days now, 60 from February, 30 from January 2028, fixed or not. What LLM-speed exploitation means for your patch SLAs

Google's bug-hunting team has scrapped the 30-day soak period it added in 2021 and will publish vulnerability details on a fixed clock because, it says, language models have collapsed the time between patch and exploit; by 2028 a bug may be public before your vendor ships a fix.

Why it matters
SecurityBy January 2028 the window between a Project Zero report and full public details is 30 days, whether or not a patch exists. If your patch SLA for browsers, operating systems and hypervisors is measured in months, it is now a published gap.
ExecutivesProject Zero's policy has set the industry norm twice before. Expect other researchers, AI bug-finding platforms and regulators to follow, and ask your vendors in the next review how fast they can ship a fix or a mitigation.
DevelopersThe team now finds several bugs a week in targets where it once found a few a year, because LLM tooling scales discovery. Feature flags, input filtering and hotpatch paths are the mechanisms Project Zero says it expects vendors to build.

Google's Project Zero published its 2026 disclosure policy on November 13, and the change is simple to state: a vulnerability it reports to a vendor becomes public X days later, with or without a fix. X is 90 today, falls to 60 on February 2, 2027, and to 30 on January 3, 2028. The team calls the scheme "Full X". Gone are the 30-day patch-adoption buffer that the 2021 "90 + 30" policy added, the 14-day grace period vendors could request, the separate rules for variants, and any link between a vendor's severity rating and the disclosure date. Whether a bug is fixed will still be tracked and stated, but it is no longer a decision point.

Why now: the model says patch-to-exploit is zero

Project Zero's post, written by Natalie Silvanovich, gives one reason for every removed element: language models. They let attackers find and exploit bugs faster, raise the odds that several parties, some hostile, find the same bug during an embargo, and let a patch be reverse-engineered into a working exploit almost immediately. The team now treats "zero patch-to-exploitation time" as a reasonable, if imperfect, model of attacker behaviour, which makes a soak period meaningless. Volume is the other driver. Project Zero says it is finding several vulnerabilities a week in targets where it once found a few a year, and a negotiation-heavy policy does not scale to that rate. The Cloud Security Alliance's note on the change adds the external data: Mandiant measured an average time-to-exploit of five days for bugs exploited in 2023, down from 32 days in 2021–22 and 63 in 2018–19, and 70% of the 138 bugs it studied were first exploited as zero-days.

What vendors argued, and what survived

Vendors consulted before publication raised three objections: a sudden switch would expose unfixed bugs while fix processes were still improving; a hard deadline would push them to fix Project Zero's reports ahead of more severe ones; and those who could not meet the clock might stop trying. The phased 90, 60, 30 schedule is the compromise, described as "challenging but realistic". Four exceptions remain. Bugs exploited in the wild fall under Google's separate seven-day policy. Bugs already public elsewhere may be published at Project Zero's discretion. A vendor can ask for earlier publication. Hardware bugs of the Spectre and Meltdown kind may get more time. The Reporting Transparency trial from July 2025, which lists the vendor, product and report date within a week without technical detail, continues; under a 30-day clock the gap between that listing and full details shrinks to roughly three weeks.

The counter-argument

The case against Full X is the one vendors made: disclosure deadlines punish the organisations with the slowest fix pipelines, which are often the ones whose users are least able to defend themselves, and a public bug with no patch helps attackers more than defenders. Project Zero's answer is that attackers armed with models are already finding these bugs, so secrecy buys less than it used to, and that vendors can protect users without a full patch through feature flags, input filtering, alternate update channels and hotpatching, the four mechanisms its October 6 post described. The CSA's Project Glasswing data cuts the other way: only 97 of 1,596 AI-discovered open-source bugs, about 6%, had confirmed patches by late May. Faster disclosure of bugs nobody is resourced to fix is the scenario to watch.

What to do

Subscribe to the Reporting Transparency list and map each new entry against your software inventory, with the expected publication date as a field. Reset patch SLAs for the families Project Zero targets, operating systems, browsers, mobile platforms, hypervisors and common libraries, so internet-exposed and high-value systems are patched in days. Pre-approve compensating controls and an emergency change path for the case where details are public and no patch exists. Weight exploitability and exposure above vendor severity, which measures impact, not the chance of discovery. Add two questions to vendor risk reviews: what non-patch mitigations can you ship, and what is your median time-to-fix for externally reported bugs. CISA's BOD 26-04 three-day tier and the EU Cyber Resilience Act's 24-hour early-warning duty, in force since September 11, already assume this tempo.

Key terms in this story

  • Full X disclosure (Project Zero) Google Project Zero's 2026 policy: a vulnerability it reports is published X days later whether or not a fix exists; X is 90 now, 60 from February 2027 and 30 from January 2028.
All 99 terms →
securitypolicyresearchGoogle
Written in our own words from the linked sources · updated 2026-11-16 · corrections: corrections@aibriefing.io

Get the brief in your inbox

One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.

Published daily by AI Briefing. We never share your address.
Like this? Get tomorrow's brief by email, free.