Anthropic introduces Cyber Mission to help protect critical infrastructure and open source with Claude
IA en un minuto newsroom · Editor: Jon Elgezabal

In 30 seconds
Anthropic is kicking off Cyber Mission, a long-haul plan to strengthen cyber defense with Claude. Its first piece brings together 11 companies, among them CrowdStrike and Rockwell Automation, to shield the power grid, water and transportation alongside Anthropic engineers. OSS Scanner, in turn, will give open-source projects that sign up periodic reviews at no cost.
Anthropic has unveiled the Anthropic Cyber Mission, which it describes as a long-term commitment to help secure the systems that everyone depends on.
Its purpose is to back defenders with tooling, research and resources, and it begins on two fronts: critical infrastructure and open-source software. For the first, the company is launching the Critical Infrastructure Defense Program, which sends frontier Claude models, engineers working on site and threat research to the providers that operators lean on to protect operational technology in power grids, water systems and transport. The program has 11 founding partners: Accenture, CrowdStrike, Dragos, Hitachi, Nozomi Networks and PwC, together with Booz Allen, Deloitte, Insane Cyber, Palo Alto Networks and Rockwell Automation. Several of them, the company says, already use Claude to fix vulnerabilities. For now it will begin by working with a small group of providers to find out which strategies work best, and firms that make security products or services aimed at critical infrastructure can put their names forward for when the program expands.
Operational technology, Anthropic explains, covers industrial networks, control software and controllers designed to last for decades. Such equipment often cannot be disconnected for patching, so known vulnerabilities stay open for years. The company had already launched a cyber defense program in June for governments at state, local, tribal and territorial level in the United States, and since then more than half of the country's states have been offered its models and technical support.
The second front is OSS Scanner, a free service that gives open-source projects recurring security scans run on Anthropic's most powerful models. Nearly all software depends on open-source code, a large part of which is kept up by small volunteer teams. Each project decides whether to join; the service takes its cue from Google's OSS-Fuzz and is aimed at projects able to handle whatever it turns up. A report contains a proof of concept showing how the flaw might be exploited, an explanation and, when there is one, a proposed fix. Because the model produces the reports and they go out with no human review, they arrive sooner, though some will contain mistakes, a miscalculated severity for instance. Anthropic expects a true-positive rate of more than 90%. Projects unable to keep pace will go on receiving disclosures verified by people, and the Defender Advantage Fund, which the company set up in August, keeps the service free.
In the company's view, finding vulnerabilities has never been easier, whereas verifying them, ranking them by priority and fixing them is still hard. That conclusion comes from Project Glasswing, in which its partners found many vulnerabilities without yet achieving a sufficient reduction in risk. This week Anthropic folded the project into its Cyber Verification Program, which has been expanded so that many more defenders can use its most capable models. Anthropic expects AI to favor the defense two years from now, though it warns that this may not be the case in the short term.
Why it matters · analysis and opinion
The hard part is no longer finding flaws but getting someone to fix them in time, and that is the bottleneck Anthropic itself acknowledges. By relying on consultancies and manufacturers that already work inside plants and industrial networks, the program reaches operators through a channel they already trust, although for now with a small group and the expansion planned for the coming months. For anyone maintaining an open-source project, OSS Scanner only helps if there are hands to review what arrives, because the reports are not checked by people and may contain errors. A company that depends on open libraries can start by knowing which ones it uses and whether their maintainers have signed up, because when a patch comes out, applying it will be up to the company.
Official source: Anthropic · Written with the help of AI: how we make the news


