Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarAnthropic Research→ original

Anthropic Shortens Secret Vulnerability Window From 90 to 30 Days Because of AI

When Anthropic finds a vulnerability in someone else's code, it reports it privately to the author, waits for a patch, and then publishes the details. That "quiet window" used to last 90 days. Now it is 30 days for serious and critical bugs and 45 days for everything else. The reason is AI: models find more holes, and it is increasingly likely that another model will find the same bug. The rule takes effect on January 1, 2027, so maintainers should get their security inboxes in order now.

AI-processed from Anthropic Research; edited by Hamidun News
Anthropic Shortens Secret Vulnerability Window From 90 to 30 Days Because of AI
Source: Anthropic Research. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceAnthropic Research ↗
Editor-in-Chief
Key metric #1If a maintainer stays silent, an external coordinator will be brought in after 15 days, not 30
Key metric #2Full technical details will be published 30 days after the patch, not 45
Key metric #3For vulnerabilities under active attack, the deadline stays at 7 days

Imagine a researcher finds a hole in a popular library. They don't shout about it across the internet. They quietly write to the author and give them time to fix it.

Only then do the details go public. This period is called the disclosure window, and for years the standard was 90 days. Now Anthropic is shortening it: to 30 days for high- and critical-severity vulnerabilities, and to 45 days for everything else.

The reason is that AI has sped up both the discovery of bugs and their rediscovery, so keeping a secret for three months has become too risky. If you maintain an open-source project or build products on third-party libraries, the gap between a report and its publication is shrinking for you.

What Exactly Is Changing

The new deadlines apply to vulnerabilities that Anthropic reports to vendors and maintainers starting January 1, 2027. Reports that are already in progress by that date will keep their existing deadlines.

There are two more changes. First, if a maintainer does not respond to the first message, Anthropic will seek to bring in an external vulnerability coordinator after 15 days instead of 30. Second, after a patch is released, the company will wait 30 days instead of 45 before publishing full technical details. The other provisions of the policy remain unchanged.

There is also a tail of older reports: serious and critical vulnerabilities that Anthropic reported more than 90 days ago and that remain undisclosed. The company is notifying the authors of these projects that they have another 30 days. After that, the reports will appear on its dashboard under the updated rules, starting January 1, 2027.

Why 90 Days No Longer Works

The old norms rested on two assumptions. First, fixes are made at human speed. Second, it is unlikely that someone will independently find the same bug. According to Anthropic, neither holds true today. Models already help write patches, and a bug found by one model is increasingly likely to be found by another. The company notes a rise in duplicate discoveries as such systems grow more capable.

An everyday analogy: a faulty lock used to go unnoticed for months because there were few lockpicking specialists. Now a multitude of automated assistants can check locks, and the chance that someone with bad intentions spots the weakness before you has grown considerably. A secret designed to last three months may not survive even until the patch.

Exceptions and Support for Maintainers

Anthropic acknowledges that not all flaws can be closed within 30 to 45 days. Programs that are hard to update (critical infrastructure, for example) and bugs that require changes to a specification or RFC are still negotiated with the vendor on a case-by-case basis. The seven-day deadline for vulnerabilities that are already being actively exploited does not change. The policy does not cover AI findings that have not been verified by a human, such as results from OSS Scanner.

The company recognizes that compressed timelines put pressure on maintainers, who are often volunteers. For that reason, each report will come with practical recommendations and, where possible, a patch or a way to reduce risk. Before sending a large number of findings to a single project, Anthropic promises to agree on a manageable pace with the maintainers.

Three programs are offered for support:

  • OSS Scanner: a free service for maintainers of critical open-source projects. It periodically scans their code with the company's most powerful models and sends findings directly to them;
  • Claude for Open Source Program: six months of Claude Ultra for free for eligible maintainers and contributors;
  • Cyber Verification Program: for security professionals, including maintainers. Claude's protective restrictions can be configured so they do not interfere with legitimate defensive work.

What to Do in Practice

If you maintain an open-source project, decide in advance who reads the security inbox, how quickly you confirm receipt of a report, and how you build a patch. Silence now costs more: after 15 days, the matter may be handed to an external coordinator. Teams that depend on third-party libraries should speed up dependency updates, since vulnerability details will become public just 30 days after a patch.

There is a caveat. The policy describes the intentions of one company and does not require other researchers to act the same way, and whether maintainers can keep up with the new pace will only become clear after January 1, 2027. But the direction is clear: as AI accelerates both the finding and the fixing of holes, the familiar slow processes will have to be rebuilt.

Story

Key Questions Answered

What Exactly Is Changing?

The new deadlines apply to vulnerabilities that Anthropic reports to vendors and maintainers starting January 1, 2027. Reports that are already in progress by that date will keep their existing deadlines. There are two more changes. First, if a maintainer does not respond to the first message, Anthropic will seek to bring in an external vulnerability coordinator after 15 days instead of 30. Second, after a patch is released, the company will wait 30 days instead of 45 before publishing full technical details.

Why 90 Days No Longer Works?

The old norms rested on two assumptions. First, fixes are made at human speed. Second, it is unlikely that someone will independently find the same bug. According to Anthropic, neither holds true today. Models already help write patches, and a bug found by one model is increasingly likely to be found by another. The company notes a rise in duplicate discoveries as such systems grow more capable. An everyday analogy: a faulty lock used to go unnoticed for months because there were few lockpicking specialists.

What are the key details on Exceptions and Support for Maintainers?

Anthropic acknowledges that not all flaws can be closed within 30 to 45 days. Programs that are hard to update (critical infrastructure, for example) and bugs that require changes to a specification or RFC are still negotiated with the vendor on a case-by-case basis. The seven-day deadline for vulnerabilities that are already being actively exploited does not change. The policy does not cover AI findings that have not been verified by a human, such as results from OSS Scanner.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…