Anthropic has announced the Anthropic Cyber Mission, a programme aimed at defenders rather than at its own model safety. It starts with two pieces. A Critical Infrastructure Defence Programme gives selected security providers access to frontier Claude models, on-site engineers and threat research, focused on the operational technology behind power grids, water systems, transport networks and government systems. Named founding partners include Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation.
The second piece is OSS Scanner, an opt-in service offering free, repeated security scans of open-source projects. Reports can include a proof of concept, an explanation and a suggested fix, and Anthropic says they are sent without human review. It expects the true-positive rate to be above 90%.
That last number is the one to watch. A scanner that files findings straight to volunteer maintainers is useful at a high true-positive rate and a nuisance at a low one, and the figure is Anthropic’s own expectation rather than a measured result. No methodology or sample is given for it.
Anthropic also folded Project Glasswing into an expanded Cyber Verification Programme, and says a June programme for state, local, tribal and territorial governments has since reached more than half of US states.
Source: Anthropic, 8 October 2026.