Anthropic Cuts Vulnerability Disclosure Window to 30 Days as AI Accelerates Bug Discovery
Anthropic announced on November 16 that it will shorten its coordinated vulnerability disclosure (CVD) policy from a flat 90-day window to 30 days for high- and critical-severity issues and 45 days for all others, effective for reports filed on or after January 1, 2027 [1][8]. The change arrives three days after Google Project Zero unveiled its own phased move to a 30-day deadline [3][6], and less than a week before Anthropic's expected Nasdaq listing [7].
New deadlines and adjacent clocks
Under the revised policy, high- and critical-severity vulnerabilities must be disclosed 30 days after reporting, down from 90. All other severities move to 45 days. Reports already in progress keep their existing deadlines [1][2][8].
Two related timelines also compress. If a maintainer does not respond to an initial report, Anthropic will escalate to an external vulnerability coordinator after 15 days instead of 30. Once a patch ships, the company will wait 30 days rather than 45 before publishing full technical details [1][4][8]. The seven-day disclosure window for vulnerabilities under active exploitation remains unchanged [1][4][8].
Rationale: AI has changed rediscovery rates
Anthropic's Frontier Red Team argues that frontier models have invalidated the assumptions behind the 90-day norm. "Models can now help develop fixes, and a bug that one model can find is increasingly likely to be found by another," the team wrote [1][8]. The company says it is seeing "a higher rate of vulnerability rediscovery as these capabilities increase" [1][8]. The old window "made sense when a vulnerability could be expected to stay private for months. We no longer believe it can" [1][2].
Google Project Zero offered a similar diagnosis, saying, "We are finding several vulnerabilities a week in targets where we once found a few vulnerabilities a year," and that attackers can reverse-engineer a patch into a working exploit quickly enough that soak time no longer protects anyone [3][6].
Exceptions and backlog handling
The policy retains case-by-case timelines for critical infrastructure and flaws requiring specification or RFC changes [1][4][8]. AI-discovered findings that have not received human review—such as output from Anthropic's OSS Scanner—are exempt from the new deadlines [1][4][8].
For high- and critical-severity issues reported more than 90 days ago that remain undisclosed, Anthropic is giving maintainers an additional 30 days' notice before publishing those reports on its dashboard starting January 1, 2027 [1][4][8].
Support for maintainers
Acknowledging the pressure on open-source maintainers, Anthropic said each report will include actionable guidance and, where possible, a patch or mitigation [1][4][8]. The company also pointed maintainers to three programs: OSS Scanner, a no-cost periodic code-scanning service for critical open-source projects; the Claude for Open Source Program, offering six months of Claude Ultra at no cost to eligible maintainers and contributors; and a Cyber Verification Program that tailors Claude's safeguards for legitimate defensive work [1][4][8].
Context: IPO timing and industry shift
The announcement comes as Anthropic prepares for a global offering of Class A common stock listed on Nasdaq under the symbol "ANT." A prospectus approved by Germany's BaFin on November 11 sets a maximum public offering price of $120 per share, with a U.S. estimated range of $70–$78, and an expected close on or about November 18, 2026 [7]. The company's governance structure concentrates voting control in a Long-Term Benefit Trust and a founder-held Class F share, limiting Class A holders to 49.9% of voting power [7].
Google Project Zero's new "Full X" policy takes a different design: it removes severity splits and post-patch waiting periods entirely, moving from 90 days today to 60 days on February 2, 2027, and 30 days on January 3, 2028 [3][6]. Project Zero argues severity reflects user impact, not attacker rediscovery likelihood [3]. Neither organization has published data on how many vendors currently fix reported bugs within 30 days [3].
What the dashboard shows
Anthropic's public CVD dashboard, updated November 2, 2026, lists 5,087 disclosed vulnerabilities across 680 open-source projects, with 629 patched and 733 identifiers (230 CVEs, 503 GHSAs) assigned [9]. The company uses Claude models, including an early snapshot of "Claude Mythos Preview," to find candidates, which are then triaged by six external security firms before disclosure [9]. The true-positive rate for manually reviewed findings stands at 94.2% [9].
What remains unknown
It is not yet clear how many vendors will meet the new deadlines, whether the compressed windows will increase the rate of public disclosure without patches, or how the industry's patching cadence will adapt. Anthropic's overdue reports, due to appear on its dashboard from January 1, 2027, will provide the first public measure of how the shorter clock works in practice [3].
Sources
- Anthropic cuts bug disclosure window to 30 days from 90 (aiweekly.co)
- Anthropic cuts its bug disclosure window to 30 days as AI ... (thenextweb.com)
- Anthropic Cuts Vulnerability Disclosure Deadline to 30 Days ... (cyberkendra.com)
- Anthropic Cuts Vulnerability Disclosure Window From 90 Days ... (finance.biggo.com)
- Google Project Zero Sets Course for 30-Day Disclosure ... (securityweek.com)
- Google Project Zero to Cut Disclosure Deadline to 30 Days ... (infosecurity-magazine.com)
- Anthropic, PBC - FWP - EEA Prospectus (sec.gov)
- Updating our coordinated vulnerability disclosure policy timelines \ Anthropic (anthropic.com)
- Anthropic's coordinated vulnerability disclosure dashboard (red.anthropic.com)