Anthropic launches Cyber Mission: frontier models and on-site engineers go to grid and open-source defenders

Defenders of critical infrastructure and open-source software gain a new channel for model access and engineering support, not just another vulnerability report. These defenders have decades of security experience but have faced severe resource shortages, and although Project Glasswing partners uncovered many vulnerabilities, cyber risk has not yet been sufficiently reduced. Anthropic says that in Glasswing it often saw months pass between a vulnerability being found and being fixed, and that in operational technology a fix may have to wait until it can be applied safely to running machinery, in some rare cases for decades.

The Critical Infrastructure Defense Program hands frontier Claude models, on-site engineers and threat research directly to trusted providers of operational technology. Power grids, water utilities, factories and transport networks run on controllers, control software and industrial networks built to last decades; these systems often cannot be taken offline to patch, so known vulnerabilities can stay unresolved for years. Founding partners are Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation; Anthropic says several partners are currently working with Claude to fix vulnerabilities, and that its first step is to work with a small cohort of providers to learn which strategies are most effective and practical.

OSS Scanner gives open-source projects periodic free scans from Anthropic's most capable models, with each report including a proof of concept, an explanation and a suggested fix where one is available. Under Project Glasswing, Anthropic scanned hundreds of widely used open-source projects, had humans triage and review many of the potential vulnerabilities, and privately reported findings to maintainers through coordinated vulnerability disclosure. Some maintainers able to triage at scale asked for everything the models found in their project, reviewed or not; OSS Scanner is therefore opt-in, its reports are model-generated and sent without human review, and Anthropic expects a true-positive rate above 90% while acknowledging that some reports will contain inaccuracies such as a wrong severity rating.

The report covers the program structure, founding partner list and work already underway that Anthropic announced; the critical-infrastructure effort explicitly starts with a small cohort of providers, and OSS Scanner is meant for projects with the capacity to keep up with surfaced findings, while others continue to receive human-verified coordinated disclosure.

Anthropic itself says critical infrastructure is hard to defend in many ways AI cannot fix; OSS Scanner reports are sent without human review and some will contain inaccuracies, with the above-90% true-positive rate only an expectation; it forecasts that in two years AI will favor defense, but says that may not be true in the near term.