The grace period is shrinking: bugs go public faster, but patches land later

Google's Project Zero, the GNOME project and the US agency CISA are all cutting their vulnerability deadlines. Their reason: AI can turn a patch into an attack tool almost instantly. Yet the data shows companies are fixing vulnerabilities more slowly than they were a year ago. We put the new deadlines side by side and checked them against the numbers.

Analysis Nov 15, 2026 12 min read Original analysis by our editors
Illustrative image, AI-generated

What this is about

When you find a security vulnerability, you normally report it to the vendor first and then keep quiet for a while. That window gives the vendor time to build a patch and get it to users. The practice is called coordinated disclosure, and for decades 90 days was the standard. Now several key players are shortening their deadlines at the same time, and they all give the same reason: language models find bugs faster and can turn a published patch into a working attack almost immediately.

Our argument: deadlines are getting shorter, but actual fixes are taking longer. The gap between the day attackers know about a bug and the day the patch lands on your device is growing. Below we compare the new rules, set them against the measured data and show who can't keep up with 30 days.

The new rules: Project Zero, GNOME, CISA

On November 13, 2026, Google Project Zero scrapped the "90+30" model it had used since 2021. Under that model, vendors had 90 days to ship a fix. Once the fix was out, a 30-day "soak time" followed so users could install the update before details went public. Vendors could also request a 14-day grace period. The new rule is "Full X": a vulnerability is disclosed X days after it's reported, patched or not. Right now that's 90 days. From February 2, 2027, it's 60, and from January 3, 2028, just 30. The grace period and the special rule for variants of previously reported bugs are gone. Severity no longer affects the timing either. The team's reasoning: severity changes the impact of an attack, not the odds that attackers find the bug.

Project Zero explains that language models have drastically cut the time it takes to work backwards from a patch to the bug it fixes. Zero time between patch and attack is a "reasonable, if imperfect, model" of attacker behavior, the team writes. Soak time therefore only creates a "false sense of secrecy." Volume is the other factor: the team used to find a handful of bugs a year and now finds several a week. Negotiating extensions case by case simply doesn't scale anymore. Instead, vendors should shrink the patch gap, the time between a finished fix and its installation on users' devices.

GNOME has gone further. Michael Catanzaro, who has handled security reports for the desktop project since November 2020 with support from Red Hat, announced on July 20 that reports received from August 1, 2026, would be disclosed after 30 days instead of 90, or sooner once a fix is available. His reasoning:

"Non-AI reports are now fairly unusual, so it no longer makes sense to optimize for them."

Most maintainers fix bugs within one to three weeks or not at all anyway, he said. He rejected the Linux kernel's approach of making suspected AI-generated reports public immediately as "unfriendly to maintainers."

The US cybersecurity agency CISA issued Binding Operational Directive BOD 26-04 on June 10. It replaces the older directives 19-02 and 22-01 with a framework of four questions: Is the system reachable from the internet? Is the vulnerability listed in the KEV catalog, the list of vulnerabilities known to be exploited? Can the attack be fully automated? Does the attacker gain partial or full control? Depending on the answers, the deadline is three, 14 or 60 days, or "at the next upgrade," according to Tenable. For the most dangerous cases, the directive requires three days plus a forensic check for signs of compromise. Mitigation counts too. Agencies must comply by December 7, 2026. CISA itself writes that attackers' use of AI could "further compress" defenders' response time.

The deadlines at a glance

Organization Deadline Applies to Status
Google Project Zero 90 → 60 → 30 days from report, fixed Disclosure since Nov 13, 2026; 60 from Feb 2, 2027, 30 from Jan 3, 2028
GNOME Security 30 days (or sooner once fixed) Disclosure reports from Aug 1, 2026
Linux kernel security team hold fix back max. 7 days, 14 in exceptional cases Disclosure of the fix AI-found bugs "treated as public"
linux-distros list max. 14 days, preferably under 7 Embargo for distributions unchanged
CERT/CC 45 days Disclosure unchanged
Anthropic (Claude findings) 90 days (+14), 7 days if actively exploited Disclosure as of March 2026
Trend Micro ZDI 120 days Disclosure unchanged
CISA BOD 26-04 3 / 14 / 60 days / next upgrade Remediation in US agencies mandatory from Dec 7, 2026
EU Cyber Resilience Act 24 h / 72 h / 14 days after fix Reporting to authorities since Sep 11, 2026

One key point: only Project Zero and GNOME have actually shortened their deadlines. We checked ZDI's and CERT/CC's own pages, and neither has changed anything. For Germany's BSI, secondary sources cite a typical range of 45 to 90 days under its 2022 CVD guideline. Whether the BSI will adjust is unclear.

What the numbers say: the gap is widening

Metric before now Source
Avg. time from disclosure to first observed exploit 115 days (2024) 44 days (2025) Flashpoint
Median time to full remediation of KEV vulnerabilities 32 days 43 days Verizon DBIR 2026
Share of KEV vulnerabilities fully remediated 38% 26% Verizon DBIR 2026
Median time KEV vulnerabilities remain externally visible – 87 days Moody's/Bitsight
Mean time-to-exploit – −7 days Mandiant M-Trends 2026

Flashpoint's data goes back further: in 2020, the average time to first attack was still 745 days, roughly 17 times what it is today. Mandiant's negative value means attackers, on average, strike before disclosure or a patch. According to Verizon, exploiting vulnerabilities is now the most common initial access vector in data breaches at 31 percent, ahead of stolen credentials at 13 percent.

Doing the math: in 2024, defenders had a head start of about 83 days between typical remediation (32 days) and the typical first attack (115 days). In 2025, at 43 versus 44 days, that head start has all but vanished. According to the Moody's data, known exploited vulnerabilities even stay visible for a median of 87 days, roughly six weeks longer than it takes for the first attack.

Our take: These figures aren't directly comparable. Flashpoint reports an average starting at disclosure, while Verizon reports a median that reportedly starts only when the vulnerability is discovered in an organization's own network. But the direction is clear, and it matches the new deadlines: they pressure finders to disclose faster while work piles up for operators. Verizon partly attributes the increase to organizations having to handle a median of 16 KEV vulnerabilities at once, up from 11.

Who can't keep up with 30 days

Android: Chipmakers and device manufacturers stand between Google and the user. A study presented at NDSS 2024 measured an average of 140 days across all examined Samsung devices between the patch level in the Android Security Bulletin and the finished firmware, and 32 days for Xiaomi. According to a Google study of 567,000 update records, 89.7 percent of vulnerabilities on unpatched Android devices can be exploited with no user interaction and little effort. On top of that, the monthly bulletins now only include high-risk vulnerabilities, and most fixes ship quarterly. In July, the alternative Android OS GrapheneOS reported that Google now backports most internally found vulnerabilities only if they're critical, and then only to Android 16 and 17. GrapheneOS blames a flood of AI-found bugs. That's GrapheneOS's account; we have no official confirmation from Google.

Routers and embedded devices: These often have no update path that users can control themselves. With ISP-supplied routers, fixes reportedly have to pass through months-long integration and testing pipelines. Project Zero itself flagged this upstream patch gap back in 2025: the chipmaker finished the fix long ago, but device makers haven't integrated it yet. For hardware, the team still holds out the prospect of rare exceptions like Spectre and Meltdown, but not as a rule.

Linux and distributions: The kernel itself is fast; a fix can be held back for seven days at most. The chain behind it is slow. As we reported, Microsoft's AI system MDASH sent 59 kernel bug reports, 55 of them by September 1. By November 11, fixes for 38 reports had landed in the main branch. So for a third of them, there was still no fix in the mainline kernel after more than ten weeks, let alone in distributions and devices. With the 7.4-rc2 test kernel, Torvalds sees fixes outpacing new reports only in the BPF area. Most other areas aren't there yet, he says.

Volunteer projects: Here, people are the bottleneck. Catanzaro steps down on November 1, and GNOME has no successor for handling security reports. As we reported, kernel developer Lorenzo Stoakes gave up reviewer roles because, by his estimate, at least half of the patches are "AI slop," carelessly auto-generated code. Sebastian Pipping publicly pushed back on GNOME: other projects shouldn't adopt the 30 days, he argued, because they become a problem as soon as several cases arrive at once. On Hacker News, commenters called Project Zero's planned 30 days "unnecessarily harsh" and asked what happens to bugs that are inherently hard to fix. Project Zero itself reports that vendors warned hard deadlines could lead teams to prioritize by deadline instead of severity, or to give up altogether.

The EU factor: the Cyber Resilience Act

Since September 11, 2026, manufacturers of products with digital elements have had to report actively exploited vulnerabilities via ENISA's Single Reporting Platform: an early warning within 24 hours, a notification with an initial assessment within 72 hours, and a final report 14 days after a fix becomes available. For open-source maintainers, called "stewards" in the law, the obligation applies from December 11, 2027, just over three weeks before Project Zero moves to 30 days.

One detail in Annex I, Part II, No. 4 of the CRA stands out. Manufacturers must publish information on fixed vulnerabilities, but in "duly justified cases" may delay this "until users have had the opportunity to apply the patch." In other words, the law provides for exactly the soak time Project Zero just abolished. Our take: Legally, that's not a contradiction, because the CRA binds manufacturers, not the researchers who find vulnerabilities. In practice, though, it will happen more often that a European manufacturer is allowed to withhold details Project Zero has already published. If the vulnerability is then exploited, the manufacturer's 24-hour reporting clock starts ticking.

What users and admins should do now

  • Turn on automatic updates, and actually restart your browser, phone and computer after updates. On Android, check under "Security patch level" how old your patch status is.
  • Look at update commitments when buying devices: not just how long they last, but how quickly patches arrive. Replace routers that no longer get vendor support.
  • Admins: inventory first. According to Flashpoint, a complete overview of all devices and systems does more than any new AI tool. CISA's four questions work for prioritization outside US agencies too: internet-reachable, in the KEV catalog, automatable, full control possible.
  • Plan for the disclosure date. Since 2025, Project Zero has listed the vendor, product and deadline on its transparency page about a week after each report. With fixed deadlines, the disclosure date is known from day one. In October 2026 alone, 23 new entries were added there, most of them involving Apple.
  • Prepare mitigations, meaning workarounds such as disabling a service, for cases where the patch arrives after disclosure.
  • Manufacturers in the EU: Set up access to the CRA reporting platform before your first real incident. According to Help Net Security, there's no API at launch; you report through a web form.

The bigger picture

The new deadlines are an understandable response. If AI tools can analyze a patch in minutes or hours, secrecy after the patch offers little protection. And Project Zero is right that a problem is only solved once the fix is running on the user's device. But only part of this is backed by evidence: the argument based on zero patch-to-exploit time is explicitly a model, not a measurement. The measurements, meanwhile, show that remediation is getting slower.

Our take: Shorter deadlines increase pressure on those who can move fast, namely large software vendors with their own update channels. For Android device makers, routers, embedded devices and volunteer projects, they tend to widen the gap between knowledge and protection, as long as update chains stay the same. We'll see whether the bet pays off starting in February 2027, when the 60-day deadline kicks in, and by early 2028 at the latest. That's when Project Zero's 30 days and the CRA reporting obligations for open-source stewards almost coincide. Another open question is whether the AI that finds vulnerabilities will speed up fixing them just as much. Torvalds hints at that for the kernel's BPF area, but not yet for the rest.

Sources

#security#project zero#vulnerabilities#open source#cyber resilience act#patch management

Ask the AI assistant

Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.

Connecting …

The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.

Comments

What do you think? Already tried it? Write it here, no sign-up needed.

  1. Loading comments …

Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.