Anthropic Launches Cyber Mission

Ledge AIWednesday, October 14, 2026
  • •Anthropic announced a long-term effort to defend critical infrastructure and open-source software on October 8.
  • •Eleven companies, including Hitachi, joined; OSS Scanner will provide free, scheduled scans to projects that apply.
  • •Experts validated 97 candidates across 48 projects: 85 met disclosure criteria, and one was a false positive.

Anthropic announced Anthropic Cyber Mission on October 8, 2026, local time, a long-term effort to help defend critical infrastructure and open-source software (OSS). It is launching a program for critical infrastructure with 11 participating companies, including Hitachi, and OSS Scanner, which provides free, scheduled scans of OSS projects.

The Critical Infrastructure Defense Program (CIDP) will provide security companies with high-performance Claude models, engineers who offer operational support, and threat research. Initial partners include Hitachi, Accenture, CrowdStrike, Dragos and Rockwell Automation, among 11 companies. The operational technology (OT) covered by the program controls equipment used in power grids, water systems, factories and transportation networks. Anthropic will work with companies supporting equipment designed for long-term operation, which can be difficult to shut down for repairs, to address vulnerabilities.

OSS Scanner offers free, scheduled scans to OSS projects that apply to participate. High-performance models, including Claude Mythos, will look for vulnerabilities and send maintainers reproducing code, explanations and, where possible, proposed fixes. AI-generated reports go directly to maintainers without human review or prioritization. This may speed up reporting, but reports may contain errors or overstate the severity of a problem.

In an initial validation, experts reviewed 97 potential vulnerabilities rated “critical” or “high” across 48 projects. Of those, 85 met Anthropic’s vulnerability disclosure criteria. Eleven of the remaining 12 were real issues that were already known or duplicated other findings; one was a false positive. The results apply only to the 97 reviewed candidates and do not establish the accuracy of all scans. The service is intended for projects able to review and handle reports; a lead maintainer must apply, and Anthropic will check whether the project meets its participation requirements.

On October 6, Anthropic combined Project Glasswing with its existing Cyber Verification Program and set three access tiers based on defensive work. Eligible models include Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1. Cyber Mission provides model access alongside hands-on support for critical infrastructure and scanning for OSS projects.