Dev.to Security 🔐 Cybersecurity 👁 0 📖 6 min read

Ninety Days Became Thirty: What the Disclosure Deadline Collapse Actually Teaches

Within four days, the two most influential vulnerability research organizations in the industry rewrote the clock that governs how long a reported bug stays private. On November 13, Google Project Zero retired the 90+30

Within four days, the two most influential vulnerability research organizations in the industry rewrote the clock that governs how long a reported bug stays private. On November 13, Google Project Zero retired the 90+30 rule it had followed since 2021, setting a course toward a fixed 30-day deadline. On November 16, Anthropic's Frontier Red Team cut its coordinated disclosure window from 90 days to 30 for high- and critical-severity issues. Both name the same cause: AI has collapsed the timelines on both sides of the embargo, and a policy written for human-speed attackers now protects the wrong thing.

The details differ in ways that matter, and a field report published by security firm Intruder on November 17 - the same week - shows what the old system was already costing.

Project Zero: from "90+30" to "Full X"

The old Project Zero policy gave vendors up to 90 days to remediate, plus a 30-day "soak time" for patch rollout. The new policy, announced by security engineer Natalie Silvanovich, is called "Full X": vulnerabilities are disclosed X days after they are reported to the vendor, regardless of whether a fix exists. The transition is gradual by design. "Full 90" took effect immediately on November 13 - which in practice means the extra 30 days of soak time are already gone. On February 2, 2027, the window becomes 60 days. On January 3, 2028, it becomes 30.

Project Zero also removed most of its special cases. Grace periods and the variant policy are out. The remaining exceptions are narrow: vulnerabilities already being exploited in the wild (handled under Google's separate 7-day policy), vulnerabilities whose details are already public, and cases where the vendor itself asks for early disclosure. Rare non-disclosure exceptions, of the kind made for Spectre and Meltdown, remain possible.

Two design decisions are worth reading closely. First, "is the bug fixed?" stops being a decision point. Project Zero will still track fixedness and say so, but disclosure no longer waits on a negotiated agreement about whether a fix counts as a fix. Second, severity is rejected as a disclosure metric: it affects a bug's impact on users, not the probability attackers find it, so it has no place in the timing rule.

The reasoning is stated plainly. LLMs increase the speed at which attackers find and exploit bugs, raise the odds that multiple parties - including malicious ones - discover the same vulnerability during an embargo, and have driven patch-to-exploit time toward zero, making soak time "no longer meaningful" and potentially harmful as a false sense of secrecy. The gradual schedule was shaped by vendor feedback: an instant 30-day rule could expose users mid-transition, distort fix priorities toward Project Zero's reports over more severe ones, and push under-resourced vendors into giving up on deadlines entirely.

Anthropic: 30 days for serious bugs, with carve-outs

Anthropic's update applies to reports sent on or after January 1, 2027; anything already in progress keeps its original deadline. For high- and critical-severity issues the window drops from 90 days to 30. Everything else gets 45. Two smaller clocks also shrink: if a maintainer does not respond, Anthropic will aim to escalate to an external vulnerability coordinator after 15 days instead of 30, and once a patch exists it will wait 30 days instead of 45 before publishing full technical details. The 7-day window for vulnerabilities under active exploitation is unchanged.

Anthropic's rationale mirrors Project Zero's, in its own words: disclosure norms "assumed that fixes moved at human speed and that independent rediscovery of the same bug was unlikely," and neither assumption holds. The company says it is already seeing a higher rate of rediscovery as model capabilities grow.

The carve-outs are pragmatic. Flaws in hard-to-deploy software such as critical infrastructure, and flaws that require changing a specification or RFC, get a negotiated remediation timeline instead of a hard 30. The new policy does not apply to AI-discovered findings that have not had human review, such as those from its OSS Scanner service. And in a move with few precedents, Anthropic disclosed that it holds high- and critical-severity reports filed over 90 days ago that already exceed its current deadline; those maintainers get 30 more days before the reports go public on Anthropic's dashboard. To soften the pressure on maintainers, Anthropic commits to shipping actionable guidance and patches or mitigations where possible, pacing large batches by agreement, and points to its no-cost programs: OSS Scanner, Claude for Open Source, and the Cyber Verification Program.

The field report that explains why

Intruder's November 17 post, written by its Head of Security Daniel Andrew, is the practitioner's counterpart to the two policy announcements, and it supplies the anecdotes the policies abstract away. In August, Intruder found a vulnerability in Sophos UTM that gives an unauthenticated attacker root by sending the device emails. Following Sophos's documented process, it took 45 days and three emails to get a first reply. A parallel bug bounty submission on Bugcrowd - filed with full details and a working exploit - was closed as N/A three weeks later, without being passed to the vendor. When contact finally happened, the explanation was volume: the vendor is inundated with reports, much of it low-quality AI-generated submissions, and the real one got lost.

The second half of the post is the part to sit with. Intruder tested the assumption behind post-patch disclosure windows on its own bug: given the public advisory and the patch file, an LLM produced a full working exploit in under an hour. If that generalizes - and Project Zero's "zero patch-to-exploitation time is a reasonable, though imperfect, model of attacker behavior" says Google's team believes it does - then the traditional post-patch grace window no longer withholds exploits from attackers, who can build their own in an afternoon. What it withholds is the technical detail defenders need to assess exposure and hunt in logs. Intruder has moved to publishing about two weeks after a patch and says it will likely compress to seven days or less. The post also claims CVE volume in Q3 2026 was more than triple Q3 2025 - its own figure, worth attributing rather than treating as settled.

What the collapse actually teaches

The disclosure debate used to be about one clock: how long attackers need. The 2026 rewrite is about a different one: how long defenders can afford to wait, given that attackers no longer need the wait at all. When rediscovery during an embargo is likely and patch-to-exploit approaches zero, a long deadline stops buying safety and starts buying only silence.

The second lesson is where the bottleneck actually moved. Project Zero's volume problem and Intruder's Sophos story are the same problem from opposite ends: triage capacity - human review of machine-scale findings - is now the scarce resource on both the researcher's side and the vendor's. Both are designing for it: Project Zero by making its policy automatable, Anthropic by carving unreviewed AI findings out of its deadlines.

The third is for vendors and maintainers, and it is unglamorous. The policies converge on the same asks: acknowledge reports in days, do not let "out of scope for rewards" become "never reached the vendor," ship fixes fast, and close the patch gap - the time between a fix shipping and users installing it - because the embargo no longer covers it. Project Zero's phrase for the old assumption is the one to retire: secrecy after the patch is a feeling, not a control.

Ninety days was a reasonable bet when it bought safety. Thirty is what is left when it doesn't.

Sources

  1. Project Zero - "Policy and Disclosure: 2026 Edition" (primary, Nov 13, 2026): https://projectzero.google/2026/11/disclosure-policy.html
  2. Anthropic Frontier Red Team - "Updating our coordinated vulnerability disclosure policy timelines" (primary, Nov 16, 2026): https://www.anthropic.com/research/coordinated-vulnerability-disclosure-timelines
  3. Intruder - "Responsible disclosure is broken" (first-hand field report, Nov 17, 2026): https://www.intruder.io/research/responsible-disclosure-is-broken
  4. BleepingComputer - "Google Project Zero cuts bug disclosures to 30 days amid AI threats" (secondary, Nov 16, 2026): https://www.bleepingcomputer.com/news/security/google-project-zero-cuts-bug-disclosure-deadline-to-30-days-amid-ai-threats/
  5. Help Net Security - "Anthropic vulnerability disclosure deadline drops to 30 days for serious bugs" (secondary, Nov 17, 2026): https://www.helpnetsecurity.com/2026/11/17/anthropic-vulnerability-disclosure-days/
  6. SecurityWeek - "Google Project Zero Sets Course for 30-Day Disclosure Deadline" (secondary, Nov 16, 2026): https://www.securityweek.com/google-project-zero-sets-course-for-30-day-disclosure-deadline/

Policy details and quotes are from the primary posts. Intruder's exploit-timing test and CVE-volume figure are attributed to Intruder. No independent testing was performed for this article.

📰 Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.