Anthropic Launches Critical Infrastructure AI Security Program With Free Open-Source Vulnerability Scanner
Anthropic's new Cyber Mission deploys Claude AI models to security companies protecting power grids, water systems, and industrial facilities, alongside a free vulnerability scanner that has already identified thousands of flaws in open-source software.
Anthropic announced October 8 the launch of its Anthropic Cyber Mission, comprising two components: a Critical Infrastructure Defense Program that deploys Claude AI models and engineers to security companies, and OSS Scanner, a free vulnerability scanning service for open-source projects. The initiative targets operational technology in plants, substations, and water systems—infrastructure that may run for decades without the ability to take systems offline for patching.
Eleven founding partners include security vendors CrowdStrike and Palo Alto Networks, industrial OT specialists Dragos, Insane Cyber, and Nozomi Networks, consulting firms Accenture, Booz Allen Hamilton, Deloitte, and PwC, plus hardware manufacturers Hitachi and Rockwell Automation. According to Anthropic, several partners are already using Claude, though the company has not disclosed commercial terms, model access arrangements, or who covers computing costs.
Proven Accuracy in Real-World Testing
Over six months, OSS Scanner identified more than 29,000 candidate vulnerabilities in widely used software, with staff manually reviewing approximately 6,000. Nearly 5,000 unreviewed vulnerability reports with proposed patches have been distributed to maintainers who requested the full batch.
Expert penetration testers validated the system's accuracy by examining 97 critical and high-severity findings across 48 projects, confirming 85 as legitimate threats—an 87.6% accuracy rate. In a case study with wolfSSL, a cryptographic library, all but two of 74 reports proved valid, with five becoming official CVE entries.
Every report includes a self-contained reproducer. Anton Arapov of OpenSSL Corporation noted that a report with a real exploit attached is "basically job done for an engineer." Where possible, candidate patches accompany write-ups, and code history bisection shows when flaws were introduced.
Reports from the new service go directly to maintainers without human review for faster delivery. Anthropic acknowledges some will contain mistakes such as incorrect severity ratings. Projects without staff capacity to handle raw findings will continue receiving human-verified reports through Anthropic's existing disclosure process.
Operational Technology as Attack Frontier
Andrew Turner from Booz Allen Hamilton identifies operational technology as "the next frontier for autonomous AI-enabled attacks." The program specifically addresses environments where equipment built to run for decades often cannot be taken offline for patches, leaving known flaws unaddressed for years.
The program builds on Anthropic's earlier Project Glasswing, launched in April and recently folded into an expanded Cyber Verification Program, which provided early access to Claude Mythos for vulnerability detection. The company established its Defender Advantage Fund in August to finance free scanner access, and has funded the Python Software Foundation, Apache Software Foundation, Alpha-Omega, and OpenSSF through the Linux Foundation.
Potential Healthcare Applications
While the source announcement focuses on power grids, water systems, and industrial plants, the operational technology challenges described—equipment that cannot be taken offline, decades-long operational lifespans, and difficulty applying security patches—mirror conditions in healthcare facilities. Hospital infrastructure including HVAC systems, building management, and backup power shares similar constraints with the explicitly mentioned critical infrastructure sectors.
Healthcare organizations that partner with the listed security vendors or consulting firms may gain access to these AI-powered vulnerability detection capabilities, though specific healthcare applications are not mentioned in the program announcement. Any application to healthcare-specific contexts such as medical device networks, clinical systems, laboratory automation, or EHR systems would represent potential extensions rather than stated program focus. Similarly, while the program's approach could theoretically support compliance with FDA medical device cybersecurity guidance or HIPAA security requirements, these regulatory frameworks are not referenced in the source material.
Outlook and Limitations
Anthropic projects that AI will favor defenders within two years. However, the company acknowledges that finding vulnerabilities has become easier while verifying, prioritizing, and fixing them remains slow work dependent on human expertise. According to Anthropic, Project Glasswing has not yet reduced cyber risk by enough, as the cost of exploiting flaws continues to fall.
Core maintainers can enroll in OSS Scanner by submitting a pull request to an Anthropic GitHub repository. Eligibility follows the OSS-Fuzz test of "critical impact on infrastructure and user security," with decisions made case by case. More partners and sectors are expected to join the Critical Infrastructure Defense Program in coming months.
Sources
Reporting & accountability
How we maintain accuracy
Health AI Central separates reported facts, analysis, and public-source data. Readers can review our standards, request a correction, or consult the relevant research methodology before relying on a claim.