DEV Community

Cover image for Anthropic Cut Bug Disclosure to 30 Days. The 90-Day Era Is Over
Paul Lovelace
Paul Lovelace

Posted on Originally published at toolscout-e2i.pages.dev

Anthropic Cut Bug Disclosure to 30 Days. The 90-Day Era Is Over

Anthropic Cut Bug Disclosure to 30 Days. The 90-Day Era Is Over

On 16 November 2026, Anthropic's Frontier Red Team announced it is cutting its coordinated vulnerability disclosure window from 90 days to 30 days for high- and critical-severity issues, and to 45 days for everything else. The new clock starts on 1 January 2027.

I have watched disclosure timelines for years, and this one feels different from the usual policy tinkering. It is not a company getting stricter for the sake of it. It is a company admitting, in public, that the assumption the whole 90-day system rested on no longer holds: that a bug can stay private for months.

What actually changed

The headline number is 30 days, but there are four moving parts, and they matter in different ways depending on which side of a report you sit on.

The window itself. High- and critical-severity issues: 90 days down to 30. All other issues: 45 days. The new timelines apply to vulnerabilities reported to vendors and maintainers on or after 1 January 2027. Anything already in progress keeps its existing deadline, so nobody's current report suddenly gets a shorter fuse.

The escalation clock. If a maintainer does not respond to Anthropic's first report, Anthropic now escalates to an external vulnerability coordinator after 15 days instead of 30. Silence gets half the patience it used to.

The details clock. Full technical details get published 30 days after a patch ships, down from 45. The patch-then-publish gap shrinks too.

The overdue backlog. This is the part that will sting some maintainers. There are high- and critical-severity vulnerabilities Anthropic reported more than 90 days ago that still have not been disclosed publicly. Those maintainers are now getting one more 30 days of notice, and then Anthropic publishes the reports on its dashboard starting 1 January 2027. The grace period for the old backlog has a hard end date.

Two things did not change. The seven-day disclosure window for any vulnerability under active exploitation stays exactly where it was. And the new policy does not apply to AI-discovered findings that have not received human review, such as findings from Anthropic's OSS Scanner. A model spitting out raw findings is not the same as a reviewed report, and Anthropic is drawing that line clearly.

Why Anthropic says the 90-day window is dead

The reasoning in Anthropic's post is blunt, and it is worth quoting the logic rather than paraphrasing it into mush: disclosure norms were built when fixes moved at human speed and independent rediscovery of the same bug was unlikely. Neither assumption holds today.

Frontier models have become much better at discovering vulnerabilities. A flaw one model can find is increasingly likely to be found by another, and Anthropic says it is seeing a higher rate of rediscovery as capabilities improve. A 90-day window made sense when a vulnerability could be expected to stay private for months. Anthropic no longer believes it can.

That is the honest frame. This is not about Anthropic being impatient. It is about the window between "one lab knows" and "every lab knows" collapsing. The disclosure clock was always a bet about how long a secret keeps. AI just shortened the odds.

The context: Google moved first, three days earlier

Anthropic did not make this move in a vacuum. On 13 November 2026, Google Project Zero scrapped its long-standing "90 + 30" policy and replaced it with a 30-day clock of its own. Two of the biggest vulnerability-finding operations in the world landing on the same number within three days is not a coincidence. It is a consensus forming.

And Anthropic has receipts for why it believes models are this good now. Its 27 October 2026 Frontier Red Team report described an unreleased Claude model building five real-world exploit chains across Chrome, Safari, Android, Windows and Linux, and finding 62 Chrome CVEs. That is not a benchmark score. That is a model doing the actual work of a vulnerability researcher, end to end.

What this means if you maintain open source

If you maintain a project, the practical question is not whether you agree with 30 days. It is whether your current workflow can survive it.

Anthropic knows the shorter window puts more pressure on maintainers, and it is offering help alongside the deadline. Three things are on the table:

  • OSS Scanner, a no-cost service that periodically scans critical open-source projects with Anthropic's strongest models and sends findings directly to maintainers.
  • Six months of Claude Ultra at no cost through the Claude for Open Source Program, for eligible maintainers and contributors.
  • The Cyber Verification Program, which tailors Claude's safeguards to enable legitimate defensive work for security professionals, including open-source maintainers.

Anthropic also says it will keep including actionable guidance with each report, along with a patch or mitigation where possible, and that it will agree on a sustainable pace with maintainers before sending a large volume of findings to a single project. That last point matters: a 30-day clock is one thing; thirty reports landing on the same Tuesday is another.

There is also a carve-out worth knowing about. Some vulnerabilities genuinely cannot be fixed quickly: flaws in hard-to-deploy software like critical infrastructure, and flaws that require changes to a specification or RFC. In those cases Anthropic says it will work with the vendor on a remediation timeline rather than holding it to the 30-day clock. The policy has a pressure valve. Use it early, not on day 29.

What to do before 1 January 2027

The new timelines only apply to reports made on or after 1 January 2027, which gives everyone about six weeks. Here is what I would actually do with them.

If you maintain a project: check whether you have any outstanding Anthropic reports older than 90 days. If you do, the 30-day notice is already ticking. Triage those first. Then look at your response process: who gets the report, who decides severity, who can ship a patch. If the answer to any of those is "whoever is free that week," a 15-day escalation clock will find that out for you.

If you do security research: the economics of your work just changed. A 30-day window means the value of a finding decays faster, and it means coordinated disclosure with multiple vendors gets harder to choreograph. Plan your reporting so the clock starts when you are ready, not before.

If you build with AI tools: nothing breaks tomorrow. But the direction is clear. The same models you use to write code are getting better at finding the holes in it, and the industry's disclosure norms are compressing to match. The 90-day era was a human-speed artifact. It is ending because the finding is no longer human-speed.

What this is not

Some honest framing, because this story is easy to overread. This is a policy change, not a product launch. There is no new model here, no new tool, no new capability to try. Anthropic is changing how fast it talks about what its models already find.

It is also not a unilateral move. Google Project Zero moved first, three days earlier, to the same 30-day number. When the two biggest finders agree on a number in the same week, that number is the new normal, whether the rest of the ecosystem has caught up or not.

And it is not a punishment of maintainers. Anthropic's post spends as much time on the support programs as on the deadline, and the carve-out for hard-to-fix flaws is real. The message reads less like "hurry up" and more like "the world changed, here is the new clock, and here is help keeping up with it."

Our take

I think Anthropic is right about the underlying fact and the industry knows it. The rediscovery argument is hard to argue with: if one frontier model can find a bug, another one probably can too, and the gap between those two discoveries is shrinking fast. A disclosure window is a bet on secrecy, and the bet has gone bad.

But the honest caveat is the other side of that bet. Thirty days is a tight window for a volunteer-maintained project, for a team with one person who can review security patches, for a codebase where a fix means coordinating a release across downstream consumers. Anthropic's support programs help, but they do not hire the maintainer. The next year will show whether 30 days is the right number or just the first number in a race downward.

One thing I am confident about: the 90-day window is not coming back. Google and Anthropic have both walked away from it in the same week. If you maintain software, your disclosure clock is now 30 days. Plan like it.

Sources: Anthropic — "Updating our coordinated vulnerability disclosure policy timelines" (https://www.anthropic.com/research/coordinated-vulnerability-disclosure-timelines), published 16 November 2026, updated 14:32 UTC; AI Weekly — "Anthropic cuts bug disclosure window to 30 days from 90" (https://aiweekly.co/alerts/anthropic-cuts-bug-disclosure-window-to-30-days-from-90); CyberKendra — "Anthropic vulnerability disclosure policy 30 days" (https://www.cyberkendra.com/2026/11/anthropic-vulnerability-disclosure-policy-30-days.html). Context: Google Project Zero's 13 November 2026 move to a 30-day clock; Anthropic's 27 October 2026 Frontier Red Team report (five exploit chains across Chrome, Safari, Android, Windows and Linux; 62 Chrome CVEs). All policy details are Anthropic's as published (vendor-reported). No hands-on testing was performed for this article. This breaking article is not the daily buyer-intent post, and it contains no affiliate links. AI disclosure: this article was drafted with AI assistance and checked against Anthropic's announcement.

Top comments (0)