Starting January 1, 2027, Anthropic gives vendors 30 days to fix high- and critical-severity vulnerabilities it reports before it discloses them, down from 90, and 45 days for everything else. It says AI models now find and rediscover bugs too fast for the old norm.
Anthropic is cutting the time vendors get to fix the security flaws it reports. In a policy update published November 16, the company says high- and critical-severity vulnerabilities it finds will be disclosed 30 days after it reports them, down from the 90 days that has been the industry norm. All other issues get 45 days.
If you maintain software Anthropic’s models might scan, or you run a vendor’s security inbox, plan for a one-month fix cycle on serious bugs. If you only use the software, expect more patches to ship faster and with shorter warning before the details go public.
What changes, and when
- Effective date. The new deadlines apply to vulnerabilities reported on or after January 1, 2027. Reports already open on that date keep their current deadlines.
- Escalation. If a maintainer does not respond, Anthropic will aim to bring in an outside vulnerability coordinator after 15 days instead of 30.
- After the patch. Full technical details come out 30 days after a fix is available, down from 45.
- Old reports. High- and critical-severity reports older than 90 days that are still undisclosed get 30 more days from notification, then go on Anthropic’s public dashboard.
- Unchanged. The 7-day window for bugs under active exploitation stays the same.
- Exceptions. Software that is hard to deploy, such as critical infrastructure, and flaws that need a specification or RFC change can get a longer timeline negotiated with the vendor. Findings from Anthropic’s automated OSS Scanner that no human has reviewed are not covered by the new deadlines.
Anthropic’s reasoning: frontier models have become much better at finding vulnerabilities, and the company says it is “seeing a higher rate of vulnerability rediscovery.” The old 90-day norm assumed fixes happened at human speed and that a second party was unlikely to find the same bug in the meantime. Anthropic says neither holds now. It gave no numbers on how many bugs its models have found or rediscovered.
For maintainers, Anthropic points to three programs: OSS Scanner (free scanning for critical open source projects), the Claude for Open Source Program (six months of Claude Ultra at no cost for eligible maintainers and contributors) and its Cyber Verification Program for defensive security work. It also says it will agree on a pace with a project before sending it a large batch of findings.
Small vendors lose their buffer first
This is our analysis. A 90-day deadline gave a small vendor or a volunteer-run open source project room to triage, fix and ship across a quarter. Thirty days fits roughly one release cycle. Large vendors with a dedicated security team will cope. A two-person hardware startup that ships firmware for a dock, a NAS or a smart display is the one that has to change how it works.
The move also fits what Google reported the same day. Its threat intelligence group found attackers using AI agents to turn new patches into working exploits with no human involved. If attackers can weaponize a fix within days, a long private window protects less, and a fast patch cycle on your side matters more than ever.
What to do: if your company ships software or firmware, check that your security contact address is published and monitored, and that you can push a fix within 30 days. On the customer side, turn on automatic updates for internet-facing tools and devices, and keep an eye on the vendors that took months to patch in the past.
Related: the NoisyTCP Linux kernel flaw, the backdoored Claude Code extension spread through Cursor and Microsoft’s November Patch Tuesday.
