The time software maintainers get before Anthropic publicly discloses serious security flaws it reports will fall from 90 days to 30, a shorter deadline the company said reflects AI’s growing ability to discover vulnerabilities.[1] The announced policy covers high- and critical-severity flaws reported from January 1, 2027; other flaws get 45 days.[1]

The trade-off is between giving maintainers time to repair software and protecting users before another model uncovers the same weakness: Anthropic said rediscovery is becoming more frequent.[1] Anthropic acknowledged the added pressure on maintainers and said models can also help develop fixes.[1]

Anthropic said it will aim to involve an outside vulnerability coordinator after 15 days without a response, down from 30, and release full technical details 30 days after a patch becomes available, down from 45.[1] Reports already in progress retain their deadlines.[1]

Anthropic said older disclosure norms assumed fixes moved at human speed and independent discovery of the same bug was unlikely.[1] The company is also giving maintainers another 30 days for serious flaws reported more than 90 days ago that remain undisclosed, before publishing those reports starting January 1, 2027.[1] Google’s Project Zero is also shortening its window, to 60 days from February 2, 2027, then 30 days from January 3, 2028, SecurityWeek reported, according to TNW.[2]

Anthropic said critical infrastructure and flaws requiring changes to technical specifications may need negotiated repair timelines; its seven-day window for flaws under active exploitation stays unchanged.[1] Reports will continue to include guidance and, where possible, a patch or mitigation.[1]

AI findings without human review are excluded, Anthropic said.[1] Help Net Security said Anthropic did not specify which deadline those findings follow.[3]