Quick Facts

  • Anthropic reported 6,157 vulnerabilities across 591 open-source projects, with only 516 patched upstream as of October 2, 2026.
  • The Cyber Mission launches two programs: the Critical Infrastructure Defense Program, covering power grids and water utilities, and the OSS Scanner, a free vulnerability discovery service for open-source maintainers.
  • Six outside security firms reviewed 6,123 findings and confirmed 5,674 as valid, a 92.7% true-positive rate.

Anthropic announced its Cyber Mission on October 8, 2026, targeting open-source software security and critical infrastructure defense with Claude models, on-site engineers, and threat research.

The initiative covers findings discovered between November 1, 2025, and October 2, 2026. Claude models produced 29,439 raw findings during that period. Of the 6,157 reported to maintainers, 5,103 were acknowledged and 516 were patched upstream, roughly 8% of disclosed issues.

Anthropic issued 584 identifiers tied to those findings: 219 CVE records and 365 GitHub Security Advisories. A single finding may carry both.

Two Programs, Two Targets

The Critical Infrastructure Defense Program sends Claude models, engineers, and threat research to 11 partner organizations: Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC, and Rockwell Automation.

The program focuses on operational technology supporting power grids, water utilities, factories, and transportation networks. These systems often cannot be taken offline for patching, and improper updates can disrupt essential services.

Dan Gunter, CEO of Insane Cyber, described the problem the program addresses: "The sites that need protection most, like remote substations, offshore platforms, and air-gapped plants, are exactly where traditional tools can't reach and specialists can't be."

Tony Baker, VP and Chief Product Security Officer at Rockwell Automation, said the technology must be "applied responsibly, validated rigorously, and deployed with safety and reliability at the forefront."

OSS Scanner: No Human Review, High Accuracy

The OSS Scanner sends periodic security reports to enrolled open-source projects without human review before delivery. Each report includes a proof of concept, an explanation of the vulnerability, and a suggested fix where available.

Anthropic targets a true-positive rate above 90% and described the scanner as inspired by Google's OSS-Fuzz. Maintainers enroll by submitting a pull request to Anthropic's public oss-scanner GitHub repository. Eligible projects must have critical impact on infrastructure or user security.

In an initial evaluation, penetration testers reviewed 97 critical and high-severity findings across 48 projects. Of those, 85 met criteria for coordinated disclosure, 11 were real but duplicate issues, and one was deemed invalid.

Open-source maintainers offered direct assessments of the scanner's output. Todd Ouska of wolfSSL said that of 74 reports received, all but two were valid and five became CVEs. OpenSSL Corporation's Anton Arapov said reports, including raw model output, were "as good as or better than what it gets from people." PostgreSQL's Noah Misch noted that several reports came with fixes his team could use nearly as-is.

Built on Project Glasswing

The Cyber Mission follows Project Glasswing, announced April 7, 2026, with partners including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, and Google. Glasswing demonstrated that frontier AI models could identify large volumes of previously undetected vulnerabilities but did not achieve a sufficient reduction in overall cyber risk.

Anthropic merged Glasswing into its expanded Cyber Verification Program, which gives qualifying security professionals access to more capable models with reduced blocking classifiers for defensive work. An earlier government-focused effort reached more than half of all U.S. states, the company said.

Anthropic acknowledged a core tension in the work: "Critical infrastructure is hard to defend in many ways that AI cannot fix."

Read more: Anthropic's Cyber Mission starts with 6,157 findings reported to maintainers and 516 patched

ai agentsanthropiccybersecurityinfrastructureopen source

This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.