Anthropic launches Cyber Mission for infrastructure and open-source defense

Anthropic’s new program pairs frontier models and on-site engineers with critical-infrastructure security providers, while a free opt-in scanner sends open-source maintainers unreviewed AI vulnerability reports.

2 min read

Illustrative close-up of a circuit board representing AI agent software infrastructure

Anthropic launched its Cyber Mission on October 8, 2026, combining two defensive initiatives: a program for critical-infrastructure security providers and a free vulnerability-scanning service for eligible open-source projects.

The Critical Infrastructure Defense Program gives participating security companies access to frontier Claude models, Anthropic threat research and on-site engineers. Its founding partners include Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation. Anthropic says several partners are already using Claude to help identify and fix weaknesses in operational-technology environments such as power, water and transportation systems.

The second initiative, OSS Scanner, lets core maintainers of eligible projects opt into periodic scans at no cost. Projects enroll through a public configuration repository and must provide a build environment that lets an offline agent audit the code.

The important limitation is explicit: OSS Scanner’s fast-track reports are generated entirely by models and arrive without human review or triage. Anthropic says that can speed disclosure, but reports may be incorrect, overstate severity or misunderstand a project’s threat model. The company recommends the service for projects that already have capacity to handle verified high- and critical-severity findings; its slower coordinated-disclosure process retains human review.

Independent reports from The Verge and Axios confirm the launch and emphasize the operational tradeoffs. Axios notes that Anthropic has not said whether infrastructure partners receive free model access or who pays computing costs, while safely testing patches on systems that cannot easily be shut down remains unresolved.

The programs expand access to AI-assisted defense, but their practical value will depend on validation quality, maintainer workload and whether recommended fixes can be deployed without disrupting essential services.

Anthropicopen-source securityOSS ScannerClaudecybersecuritycritical infrastructurevulnerability management