InformationalVulnerabilityMonitoring

CSO Online reports Google and Anthropic shorten disclosure deadlines

CSO Online reports Project Zero and Anthropic are shortening bug disclosure deadlines; check how fast your team can apply patches or mitigations.

Published
Exploitation
No specific vulnerability is involved; CSO Online reports Google’s Threat Intelligence Group documented an automated discovery-and-exploitation framework called exploit_agent in the wild

What happened

According to CSO Online, Google Project Zero says it is cutting its vulnerability disclosure deadline from 90 days to 60 days starting in February, and to 30 days starting in January 2028. CSO Online reports that Anthropic then announced, on the following Monday, that it will move from 90 days to 30 days for high- and critical-severity issues and to 45 days for all other issues, starting in January. The outlet adds that other security companies and researchers “will likely follow”, which is its own expectation, not a confirmed fact.

CSO Online quotes Project Zero researcher Natalie Silvanovich as saying that large language models have increased the speed at which attackers can find and exploit vulnerabilities, and the risk that several parties find the same bug. Anthropic’s Frontier Red Team is quoted as saying a 90-day window no longer fits, because a vulnerability can no longer be expected to stay private for months.

According to CSO Online, other changes in Project Zero’s policy are:

  • The 30-day “soak time” grace period, part of a “90+30” policy since 2021, is dropped effective immediately.
  • Severity is no longer used as a disclosure metric, because Silvanovich says attackers are as likely to find low-risk flaws as high-risk ones.
  • The team will no longer check whether a vendor’s patch fixes the issue before deciding to release details.

CSO Online also reports that Google’s Threat Intelligence Group (GTIG) documented what it calls the first fully automated AI-powered vulnerability discovery-and-exploitation pipeline it has observed in the wild, named exploit_agent. Per the article, it analyzes code changes after security patches and generates exploits without an operator directing each step. The article does not say which software was targeted or how widely it was used, so that is not confirmed.

CSO Online cites further figures from third parties, none independently verified here:

  • The number of CVEs issued this year exceeded 100,000, more than double last year’s count.
  • Security firm XOR’s benchmark found the best models produced complete fixes for about a third of the flaws: 35.3% for GPT-6 Astra and 28.6% for Claude Fable 5.1. Across all systems, 11.9% of patch attempts earned full credit.
  • According to a VulnCheck analysis cited by CSO Online, Anthropic’s public ledger counts 37,966 findings, but only 217 of the 7,349 flaws made public so far have been fixed.

Am I affected?

This is a change in disclosure policy, not a single vulnerability, and the source names no CVE, product or fixed version. It mainly concerns software vendors and open-source maintainers who receive reports from Project Zero or Anthropic, and the teams who depend on that software. CSO Online reports that details may now become public sooner, in some cases before a fix has been verified or is available. Whether a specific product you run is affected is not confirmed by the source.

What to do now

  1. Review how long your team takes to move from a vendor or upstream advisory to a deployed patch. CSO Online reports the new deadlines are as short as 30 days.
  2. Prepare a process for the case where disclosure arrives before a fix. ArmorCode’s Ramy Rahman tells CSO that teams need a way to evaluate exposure, apply available mitigations and track remediation.
  3. Identify which compensating controls could reduce risk when a flaw is disclosed. Rahman says organizations should be better prepared to assess this.
  4. Consider hardening systems at deployment time and removing software you do not use. Rahman says leaner deployments can reduce exposure and make findings easier to manage.
  5. If you maintain software, note that Anthropic says it will include actionable guidance with each report, and a patch or mitigation where possible. CSO Online also notes that AI-generated patches are often reported as incomplete, so review any proposed fix carefully.

Sources