One of the four high-severity bugs let workspace users read Kubernetes Secrets from other namespaces; GitLab.com is already patched and Dedicated customers need do nothing.
Anthropic gives maintainers 30 days, not 90, to fix serious flaws
EditorialBy TrustList Editorial
The new clock covers reports sent from 1 January 2027; older high and critical findings still unpublished after 90 days get a final 30 days before disclosure.
- United States
- California
- Cybersecurity
- Vulnerability Management
- +2 more
About Anthropic gives maintainers 30 days, not 90, to fix serious flaws
Anthropic gives maintainers 30 days, not 90, to fix serious flaws
16 November 2026: Anthropic will publish details of high and critical vulnerabilities its systems find 30 days after reporting them, and of all other flaws after 45 days, replacing the 90-day window it has used until now. The shorter deadlines apply to anything reported to a vendor or maintainer on or after 1 January 2027, and they land hardest on open source projects with few people to triage reports.
Not yet independently verified. The timelines and the figures are Anthropic's own, from its policy post and its disclosure ledger. The post does not say how many older findings fall under the final 30-day notice, and we have not seen a response from the open source projects affected. We will update this when it can be confirmed, and remove this note.
Four clocks change, one stays
The policy post sets out the new deadlines beside the old ones:
- high and critical severity: public after 30 days, down from 90;
- every other severity: public after 45 days, down from 90;
- a maintainer who does not answer the first report: escalation to an outside vulnerability coordinator after 15 days, down from 30;
- once a patch exists: full technical detail shared after 30 days, down from 45.
The seven-day window for flaws already being exploited is unchanged. Reports in progress on 1 January keep the deadlines they started with, and the policy does not cover machine-found results that no person has reviewed.
Software that is hard to update gets separate treatment. For critical infrastructure, or a flaw whose fix needs a change to a specification or an RFC, Anthropic says it will negotiate a remediation timeline.
The reason given: the next model finds the same bug
Anthropic argues that 90 days assumed fixes moved at human speed and that nobody else was likely to find the same flaw in the meantime. Its post says the second assumption no longer holds: "a bug that one model can find is increasingly likely to be found by another".
Its ledger gives the scale. As of 2 November 2026 the company had disclosed 5,087 vulnerabilities across 680 open source projects, of which 629 were patched to its knowledge, with 733 identifiers issued: 230 CVE records and 503 GitHub Security Advisories. Six outside security research firms triage the findings before they reach maintainers.
A final 30 days for the oldest reports
One part takes effect before January. For high and critical findings reported more than 90 days ago and still not public, maintainers are being told they have a further 30 days, after which Anthropic will disclose. A project with an open report of that age should assume the details will be public about a month after it receives that notice.
Two disclosure deadlines moved in the same week
Google's Project Zero announced its own cut, from 90 days to 60 from February 2027 and to 30 from January 2028, which we covered in Project Zero's new disclosure policy. Anthropic's schedule is the shorter of the two and starts sooner.
For a company that ships software built on open source, the practical change is in the supply chain: a serious flaw in a dependency may be public a month after it is reported, whether or not a fixed release exists. Teams that patch dependencies on a quarterly cycle will be exposed for longer than the disclosure window. Our catalogue lists 17 products in the vulnerability management software category, and Anthropic has a profile at Anthropic.
The first reports under the 30-day clock can be made on 1 January 2027, so the first disclosures under it fall at the end of that month.
Sources
- Anthropic: coordinated vulnerability disclosure timelines, 16 November 2026
- Anthropic: coordinated vulnerability disclosure ledger (figures to 2 November 2026), 2 November 2026
- iThome: report on the change, 17 November 2026
Categories & features
- United States
- California
- Cybersecurity
- Vulnerability Management
- Patch Management
- Artificial Intelligence - AI
TrustList Weekly
The week in software and IT, in one email
The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.
You may be interested in
A crafted email gives an unauthenticated attacker root through SQL injection in AsyncOS (CVSS 9.8). Cisco has known of attacks since September and replaced the 15.5.5-014 fix on 31 October after exploit code appeared online.
CVE-2026-111497 is rated 9.2 by Google and hits every Brotli release from 1.0.3 to 1.2.0. Envoy, the PyPI packages and Debian have shipped fixes; Ubuntu still ranks it medium.
CVE-2026-76466 scores 9.8 and needs no login: persistent cookies and weak certificate checks between cluster peers reach the database service. There is no workaround, and X14 users must migrate.
Six Critical memory bugs in Skia, Accessibility, CSS, Paint and WebRTC lead the list, with 61 High, 197 Medium and 89 Low. Google gave no sign of exploitation in the wild.
CVE-2025-61624 lets an admin with write rights create or delete files through crafted CLI arguments on FortiOS, FortiPAM, FortiProxy and FortiSwitchManager. Older branches have no fix at all.
Mozilla now issues a CVE for every memory safety bug, so the Firefox 158 advisory lists 77. Thunderbird 158, 153.5 and Firefox ESR 115.43 carry matching fixes.
One flaw sits in image optimisation of SVG files and reaches back to Next.js 10; the other hits the use cache directive in 16.x and has no workaround. A third advisory fixes a Server Actions denial of service.
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- More United StatesThe ranking for this subject
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.