Anthropic cuts bug disclosure window to 30 days as AI speeds up rediscovery
Anthropic is slashing its vulnerability disclosure window from 90 days to 30 days because AI models are finding the same bugs faster.
Policy · Source: Anthropic
What happened
Anthropic is rewriting the rules for software security. Starting January 1, 2027, the company will publicly disclose high and critical severity vulnerabilities just 30 days after reporting them to vendors. This is a massive drop from the industry standard 90 days. For all other bugs, the deadline shrinks to 45 days. The seven-day window for actively exploited bugs remains unchanged.
The reason is simple. AI models are getting too good at finding flaws. Anthropic says the old assumption that a bug could stay secret for months is dead. If one model finds a vulnerability, another model is highly likely to find it soon after. The rate of vulnerability rediscovery is increasing as AI capabilities grow. Models are also helping developers write the fixes, which speeds up the remediation process and makes the old 90-day window obsolete.
The clock is speeding up everywhere. If a maintainer ignores the initial report, Anthropic will escalate to a third-party coordinator in 15 days instead of 30. Once a patch drops, full technical details go public in 30 days instead of 45. Anthropic is also clearing its backlog. Maintainers with undisclosed critical bugs older than 90 days are getting a final 30-day notice before those reports go public.
Key facts
- 30 days — New disclosure deadline for high and critical bugs
- January 1, 2027 — Date the new disclosure timelines take effect
- 15 days — Time before Anthropic escalates an ignored report
- OSS Scanner — Free code scanning service for critical open-source projects
Why it matters
If you build or maintain software, your timeline to triage, fix, and ship patches just got cut by two-thirds. You no longer have a three-month safety net to quietly patch a critical flaw before the public finds out. You have to move at the speed of AI. This means automating your testing and deployment pipelines to handle rapid security updates. Companies that rely on manual security reviews and slow release cycles will fail to meet these new deadlines and will see their vulnerabilities exposed.
The era of delaying software updates is over for end users and enterprises. With full technical details of critical bugs hitting the public dashboard in 30 days, attackers will have a faster roadmap to exploit unpatched systems. Vulnerability rediscovery by bad actors using their own models will force the entire industry to adopt these aggressive timelines. The gap between a patch being available and an exploit being used is shrinking to zero.
For builders
Faster patch cycles are mandatory
Maintainers lose the 90-day grace period for critical bugs. If you cannot ship a fix in 30 days, Anthropic will publish the flaw on their dashboard for everyone to see. You pay the price in public trust and user security if your deployment cycle is too slow.
Free AI tools for open source
Anthropic is offering free help to soften the blow for developers. Maintainers of critical open-source projects can apply for OSS Scanner to get automated model-driven code reviews. Eligible contributors also get six months of Claude Ultra for free to help write fixes.
Exceptions for critical infrastructure
Not everything can be patched in a month. Anthropic will negotiate custom timelines for hard-to-deploy software like critical infrastructure or protocol-level RFC changes. Builders in these slow-moving sectors avoid the immediate 30-day cliff, but they still need to prove they are working on a fix.
My take
I have always delayed my software updates. That habit is now a massive security risk. If AI can find a bug, another AI will find it again, so hiding it for 90 days is pointless. Update your systems the day the patch arrives, or you will get breached.
Original reporting: Anthropic. This is my rewrite and opinion.