Anthropic launches a Cyber Mission to help secure critical infrastructure

Anthropic launched a long-term initiative it calls the Cyber Mission, centred on a Critical Infrastructure Defense Program that gives security providers working on power grids, water systems, transportation networks and government systems access to frontier Claude models, on-site Anthropic engineers and threat research. Eleven organisations signed on as founding partners, including Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Palo Alto Networks, PwC and Rockwell Automation; Anthropic said several were already using the access to fix vulnerabilities in client systems.

Palo Alto Networks’ Sam Rubin said the programme responded to attackers who were “using AI to find and exploit exposures at machine speed.” The company framed the launch as the latest step after Project Glasswing, its earlier coordinated vulnerability-scanning effort, and said more than half of US states had already been offered frontier Claude access through a state-level cyber defence programme it opened in June.

The announcement followed, two days earlier, an expansion of Anthropic’s Cyber Verification Program — the gated route by which vetted security professionals get access to Claude with reduced safety restrictions for dual-use cybersecurity work. Anthropic folded two prior schemes, Project Glasswing and the original verification programme, into one offering with three tiers: Defense Access for incident response and malware analysis, open to company security teams, universities and individual researchers with a vulnerability-disclosure record and reviewed within days; Red Team Access for authorised penetration testing, restricted to organisations and reviewed over several weeks; and Specialized Access for testing safety-critical systems such as power grids, available only with government involvement. Anthropic said partners in the Glasswing programme had surfaced more than 129,000 verified vulnerabilities between April and July.

Both moves extend a pattern among frontier labs of offering capable but access-restricted models to defenders rather than releasing them generally, on the argument that cybersecurity tooling is inherently dual-use. The scale of the commitment — eleven named partners and a tiered verification scheme covering critical infrastructure specifically — distinguishes it from the narrower pilot programmes labs ran earlier in 2026.