Anthropic launches AI security push for power grids and open-source code
Anthropic, the AI company behind Claude, on Thursday launched a program with 11 security firms to protect power grids, water systems and other key services. It also opened a tool that sends automated bug reports to major open-source projects. The company says hackers using AI will keep the upper hand for about two more years.
Anthropic, the company behind the Claude AI models, launched a new cybersecurity effort on Thursday. It calls the effort the Anthropic Cyber Mission, and it has two parts.
The first is a Critical Infrastructure Defense Program. It pairs Claude models, Anthropic engineers and threat research with 11 security companies. They include Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Palo Alto Networks, PwC and Rockwell Automation. The goal is to find and fix weak spots in systems that run power grids, water plants and factories, CyberScoop reported.
That equipment is hard to patch. It is built to run for decades and often cannot be shut off for an update, so known flaws can sit for years, SiliconANGLE reported. Anthropic said frontier models can help fix weaknesses "before those weaknesses are used to cut off power or make water unsafe."
The second part is OSS Scanner. It gives major open-source projects regular scans for security bugs. Each report explains how a flaw could be exploited and often suggests a fix. Reports go straight to maintainers without a human check, so they arrive faster but may contain mistakes, Anthropic said.
The idea grew from a backlog. Anthropic's models flagged more than 29,000 possible flaws in widely used software over six months, SiliconANGLE reported. Staff reviewed only about 6,000 by hand. Some maintainers asked for the unreviewed reports anyway.
Not every project qualifies. Anthropic is using the same test as Google's OSS-Fuzz program: projects with a critical impact on infrastructure and user security, The Register reported.
The launch comes as AI tools get better at finding and exploiting software flaws. Anthropic says AI will favor defenders within two years. Until then, it says, attackers have the upper hand.
The Register was skeptical. It compared the move to "a careless camper starting a wildfire and later funding a fire department."
What we know: Anthropic is offering its models and engineers to infrastructure defenders and automated scans to key open-source projects.
What we don't: Who pays the computing costs for partners, and how fixes will reach running utility systems.
Where outlets agree CyberScoop and SiliconANGLE both report that 11 companies, including CrowdStrike, Palo Alto Networks and Rockwell Automation, are founding partners in the infrastructure program.
Where they differ SiliconANGLE reports the OSS Scanner is free, funded by Anthropic's Defender Advantage Fund, while The Register says it is not entirely free because maintainers' inputs and outputs can be used for model training.