Anthropic will provide Claude models, engineers and funding for critical-infrastructure defense, while offering recurring scans to open-source projects.
Anthropic has launched the Anthropic Cyber Mission, a long-term effort to apply Claude models, engineering support and funding to cyber defense. Its first initiatives are the Critical Infrastructure Defense Program for trusted security providers and OSS Scanner, a free service for participating open-source projects.
The infrastructure program combines Anthropic’s frontier Claude models, on-site engineers and threat research. Its 11 founding partners are Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation. Anthropic says several are already using Claude to fix vulnerabilities and support customers, while the initial small cohort will help identify practical strategies.
The work initially targets operational technology behind power grids, water systems, factories and transport networks. Such equipment may remain in service for decades and often cannot be taken offline for patching. Anthropic acknowledges that AI cannot resolve many of the underlying difficulties: industrial systems are proprietary, modifications require specialist validation, and mistakes can interrupt physical operations.
OSS Scanner gives enrolled open-source projects periodic security scans from Anthropic’s most capable models at no charge. Each model-generated report is intended to include a proof of concept showing how a bug could be exploited, an explanation and, where available, a suggested fix. The service is designed for maintainers able to process a continuing stream of findings.
Reports are sent without human review, which shortens delivery time but can introduce inaccuracies, including incorrect severity ratings. Anthropic forecasts a true-positive rate above 90%, but presents that figure as an expectation rather than an independently validated operating result. Projects unable to triage findings at scale will continue to receive human-verified reports through the company’s coordinated vulnerability disclosure process.
The two disclosure routes create a practical trade-off. Well-resourced maintainers can receive faster, unreviewed model output and conduct their own triage; other teams can use a slower channel with human verification. In either case, finding a suspected flaw is distinct from confirming its risk, developing a safe repair and deploying that repair.
Anthropic says the initiatives build on Project Glasswing, which scanned hundreds of widely used open-source projects and used human review for many potential vulnerabilities. Glasswing has been merged into the expanded Cyber Verification Program. Over the coming months, Anthropic plans to add partners and sectors to the infrastructure program, but its announcement gives no specific expansion schedule.
| Initiative | Intended users | What it provides |
|---|---|---|
| Critical Infrastructure Defense Program | Providers protecting critical infrastructure | Claude models, on-site engineers and threat research |
| OSS Scanner | Maintainers of enrolled open-source projects | Periodic model-based scans, exploitation details and a suggested fix |
Sources
Originally published by AI Feed: Anthropic Launches Claude Programs for Cyber Defense.
Top comments (0)