Anthropic cuts serious-bug disclosure deadline to 30 days
Original title: Anthropic Cuts Disclosure Deadline to 30 Days For Serious Bugs
From January 1, 2027, high and critical findings get 30 days and other issues 45, with escalation after 15 days of silence.
Anthropic's Frontier Red Team said vendors will have less time to fix flaws its Claude models report before public disclosure. Windows that start on or after January 1, 2027 replace the prior 90-day serious-bug clock; reports already in progress keep existing deadlines. Bugs under active exploitation still use a seven-day window, and AI findings without human review, including OSS Scanner output, are excluded. Anthropic also said it is adding 30 days for high- and critical-severity bugs that already passed the current 90-day deadline before publishing them.
Unconfirmed: The exploit_agent server finding is Google's same-day report, not part of Anthropic's policy change.
- Full technical details publish 30 days after a patch, down from 45.
- Escalation to an external coordinator moves from 30 days of silence to 15.
- Hard-to-deploy software and specification changes can use a negotiated timeline.
- Project Zero plans 60 days from February 2, 2027 and 30 days from January 3, 2028.
- Google's Threat Intelligence Group separately reported an exposed exploit_agent server.
- Malware
- exploit_agent
- Organizations
- Google Project ZeroGoogle Threat Intelligence Group
- AI models
- Claude
From the original article
2 min read at infosecurity-magazine.com
Anthropic will publish serious flaws found by its AI models 30 days after reporting them
By Alessandro Mascellino · infosecurity-magazine.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.