SIAgentNews.com
← All news
DevelopingClaudeSecurity

Anthropic cuts its bug disclosure deadline for serious flaws from 90 days to 30

Anthropic's Frontier Red Team will publish serious security flaws 30 days after reporting them, because AI now finds bugs faster than people can fix them.

· First reported by infosecu.technews.tw

30 daysnew disclosure window for High and Critical flaws, down from 90

The short version

  • High and Critical flaws go public 30 days after reporting, down from 90 days
  • Other severity levels get a 45-day deadline
  • Applies to reports filed on or after January 1, 2027; older cases keep old deadlines
  • Flaws already being exploited by attackers still go public within 7 days
  • Google's Project Zero is also shortening its deadline, to 60 days and later 30

Why it matters

When AI makes finding bugs cheap, fixing them becomes the slow step, and public disclosure clocks are how pressure to fix gets applied. If you run software from others, the time between a fix being released and installed on your machines is the part you control. Expect security updates to arrive on tighter schedules.

What Anthropic changed

As infosecu.technews.tw reported, Anthropic's Frontier Red Team announced on November 16, 2026 that it is revising its coordinated vulnerability disclosure policy. That is the practice of privately telling a software maker about a flaw, then publishing details after a set waiting period.

For flaws rated High or Critical, the wait drops from 90 days to 30. Other severity levels get 45 days. The new rules cover reports made on or after January 1, 2027, while cases already in progress keep their old deadlines.

Smaller timers shrink too

Two other clocks are shortened. If a software maintainer never answers the first report, Anthropic will hand the case to an outside coordinating body after 15 days instead of 30. After a patch ships, the wait before full technical details go public falls from 45 days to 30.

Some Critical flaws reported more than 90 days ago and still not public get a final 30-day grace period for maintainers. After that, they are posted on Anthropic's official dashboard under the new policy.

Why the old rule no longer fits

Anthropic says 90 days made sense when a flaw could stay secret for months. The old approach assumed fixes move at human speed and that two researchers rarely find the same bug. With AI models, one model's discovery is increasingly likely to be found soon by another.

That raises the risk that a flaw someone has not yet fixed is already known to others, including attackers.

Exceptions and pressure on maintainers

The policy leaves room for flexibility. If a flaw touches critical infrastructure software or a foundational technology standard and needs longer to fix, Anthropic will negotiate a schedule with the vendor. Flaws already under active attack keep the 7-day emergency rule. Pure AI scan results with no human review, such as output from its OSS Scanner service, fall outside the policy.

Anthropic admits shorter deadlines may add stress for maintainers. It says every report will include a suggested fix, and it will supply a patch directly where it can.

It is not alone. Google's Project Zero will move to 60 days from February 2, 2027, then 30 days from January 3, 2028.

What to watch

  • Whether other security teams and AI labs follow Anthropic and Google with shorter deadlines
  • How software maintainers cope, and how often negotiated extensions are used
  • Whether faster disclosure leads to faster patching in practice

Try it

Turn on automatic updates for your devices and key apps, and name one person who reads security notices for your team or household.

Sources and links

First reported by infosecu.technews.tw. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news