Anthropic Says It Will Give Software Makers 30 Days, Not 90, to Fix Serious Bugs From 2027

Anthropic says that starting January 1, 2027 it will publish the high and critical severity bugs it reports after 30 days instead of 90, arguing AI has made the old disclosure calendar obsolete. Some overdue reports are about to get a final 30 day warning.

Nov 17, 2026
Anthropic Says It Will Give Software Makers 30 Days, Not 90, to Fix Serious Bugs From 2027
An hourglass silhouetted against a grey sky

Anthropic is putting software makers on a much shorter leash. In a policy update posted Nov. 16 by its Frontier Red Team, the company said it will cut the private window it gives vendors and open source maintainers to fix the bugs it reports from 90 days to 30 days for high and critical severity issues, and to 45 days for everything else. After that, the details go public.

The clock does not start yet. Anthropic said the new deadlines apply to vulnerabilities reported on or after January 1, 2027, while reports already in progress keep their existing deadlines. Two smaller clocks shrink too: if a maintainer ignores the first report, Anthropic will aim to escalate to an outside coordinator after 15 days instead of 30, and it will wait 30 days instead of 45 after a patch ships before releasing full technical details.

Then there is the part that should make some vendors sweat. Anthropic said it is holding high and critical severity reports filed more than 90 days ago that have blown past its current deadline and remain undisclosed. Those maintainers are being told they get an extra 30 days, after which Anthropic intends to publish the reports on its dashboard under the updated policy starting January 1, 2027. As Help Net Security pointed out, those flaws have already outlasted a deadline three times longer than the one Anthropic is about to adopt.

The argument is that AI broke the old calendar

Anthropic wrote that frontier models have become much better at discovering vulnerabilities over the last several months, and that it is seeing a higher rate of rediscovery, meaning different people or models stumbling onto the same flaw.

"These policies assumed that fixes moved at human speed and that independent rediscovery of the same bug was unlikely. Neither assumption holds today." Anthropic Frontier Red Team

The logic is blunt. If a bug one model can find is increasingly likely to be found by another, as Anthropic argues, then every extra week of secrecy is a week in which someone less friendly may already have it. Anthropic also said models can now help develop fixes, its answer to the complaint that 30 days is too short to patch anything complicated. That claim deserves watching, not faith. This site has already reported on Google saying it found an AI agent turning security patches into attacks with no human help, which shows how fast the window after disclosure can close too.

Anthropic is not alone in moving, though it is moving faster. Google Project Zero, the security research team inside Google, still lists 90 days as its current deadline, but its published policy says it plans to move to 60 days on February 2, 2027 and to 30 days on January 3, 2028. Anthropic would hit 30 days for serious bugs a full year before Project Zero says it will.

The policy does have escape hatches. Anthropic said flaws in hard to deploy software such as critical infrastructure, and flaws that require changes to a specification or RFC, cannot always be fixed in 30 or 45 days, and that it will work out a remediation timeline with the vendor in those cases. Its seven day window for any vulnerability under active exploitation stays unchanged. The update also skips AI discovered findings without human review, such as those from its OSS Scanner service, and Help Net Security noted Anthropic did not say which deadline those follow.

Anthropic concedes that shorter timelines may put more pressure on maintainers. In return, it said each report will keep including actionable guidance and, where possible, a patch or mitigation, and it will agree on a sustainable pace before sending a large batch of findings to one project. It is also pointing maintainers to the free OSS Scanner and to six months of Claude Ultra through its Claude for Open Source Program.

That is a carrot, but the stick is the real story. For maintainers, especially on open source projects, a lab with frontier models and a public dashboard just set the pace, and the old excuse of needing three months is running out of road.