Anthropic is shortening its disclosure deadline for flaws Claude finds: high and critical vulnerabilities go public after 30 days instead of 90 from 2027. Anthropic's reason is that a bug one AI model finds is increasingly found by another, which matters to every marketing site running plugins.
Key facts
- In an announcement on Nov 16, 2026, Anthropic cut its disclosure window for high and critical vulnerabilities from 90 to 30 days, and to 45 days for all others.
- The new windows apply to reports filed on or after January 1, 2027; actively exploited flaws stay at seven days.
- Anthropic says AI models have become much better at finding flaws and that the same bug is more often rediscovered by another model.
Researched and written with AI, to Luigi's brief, positions and editorial rules. Automatically checked against the sources. How I work →
Look at your website maintenance contract. Plugin updates are probably listed as "monthly" or "quarterly". Those rhythms come from an era when a security flaw could stay a quiet secret between the finder and the vendor for three months. Anthropic just officially ended that era.
What Anthropic is changing
Anthropic's Frontier Red Team tightened its coordinated disclosure rules on Nov 16. When Claude finds a flaw and Anthropic's people verify it, the vendor gets a report. The time the vendor has to fix it before Anthropic goes public changes like this:
- High and critical severity: 30 days instead of 90.
- Everything else: 45 days instead of 90.
- If a maintainer doesn't respond, Anthropic escalates to an outside coordinator after 15 days instead of 30.
- Full technical details after a patch: after 30 days instead of 45.
The new windows apply to reports filed on or after January 1, 2027. Cases already in progress keep their deadlines, and actively exploited flaws stay at seven days. There are exceptions for hard-to-deploy software such as critical infrastructure.
Why 90 days no longer holds
The reasoning is the real wake-up call. According to Anthropic, the old norm rested on two assumptions: fixes move at human speed, and independent rediscovery of the same bug is unlikely. Neither holds anymore. A bug one model finds is increasingly found by another, and Anthropic says it is seeing more of those rediscoveries.
Put it next to the story I published earlier today: Google has, for the first time, observed attacker AI that turns patches into exploits on its own. Defenders are shrinking the grace period, attackers are shrinking the time after the patch. The window between "flaw known" and "flaw exploited" is closing from both sides.
The catch sits in marketing, not IT
In my view this hits marketing teams harder than IT. IT has patch processes. Marketing has WordPress plugins, form tools, tracking scripts, shop apps and that 2023 campaign microsite built by an agency that no longer exists in that form. AI doesn't fix broken ways of working, it widens the cracks in the foundation. A maintenance contract on a quarterly rhythm is one of those cracks.
And marketing will own more software, not less: AI-built apps nobody runs and credentials sitting in AI logs show where the next open door is.
What I would do now
- List every website, microsite, plugin and marketing app, each with one named owner. No owner? Switch it off.
- Put a patch deadline into the agency or maintenance contract: critical updates within days, not "in the next maintenance window".
- Turn on automatic security updates where the system offers them, and when you build new AI landing pages, plan the maintenance from day one.
Bottom lineThe grace period for security flaws is shrinking to one month. Your maintenance plan has to be faster than that.
Got AI news? Submit news →
FAQ
What is Anthropic's new disclosure deadline?
30 days instead of 90 for high and critical vulnerabilities, 45 days for all others. If a maintainer doesn't respond, Anthropic escalates to an outside coordinator after 15 days.
When does the 30-day deadline start?
For reports filed on or after January 1, 2027. Cases already in progress keep their deadlines, and actively exploited flaws stay at seven days.
Why is Anthropic shortening the deadline?
Anthropic says AI models have become much better at finding flaws and the same bug is more often found by another model, so 90 days no longer reflects how long a flaw stays private.
Source
Anthropic (Frontier Red Team): Updating our coordinated vulnerability disclosure timelines




