Anthropic has introduced a new initiative called the Critical Infrastructure Defense Program, which will make the company’s most powerful AI models, on-site engineers and threat-research capabilities available to firms responsible for protecting power grids, water utilities and other critical infrastructure.
Why it matters
Operators of critical infrastructure often manage aging, complex systems and face difficulty securing them. At the same time, advances in AI can make cyberattacks faster and cheaper to carry out, increasing urgency for stronger defensive measures.
Program details
Under the program, Anthropic will provide access to its frontier AI models, on-site engineering support and threat research to participating companies. These firms can use those resources to identify and remediate vulnerabilities in the systems of their customers.
Founding partners of the program include Accenture, Booz Allen Hamilton, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation. Anthropic said several partners are already using Claude to find vulnerabilities and to assist their customers in addressing them.
OSS Scanner: free AI-powered scans for open-source projects
Anthropic is also launching OSS Scanner, a free AI-driven vulnerability scanning service for open-source software projects. Participating projects will receive periodic scans performed by Claude and automated reports that describe identified vulnerabilities, how they could be exploited, and suggested fixes.
The company cautions that these reports are sent to maintainers without human review, so some findings may be inaccurate.
Context and prior work
Anthropic and other AI developers, including OpenAI, are seeking ways to put their most capable models into the hands of cyber defenders as adversaries gain access to similar capabilities. The new initiative builds on lessons from Project Glasswing, which previously gave vetted organizations access to Anthropic’s most capable models for vulnerability discovery. Anthropic says the project showed how quickly AI can surface vulnerabilities while also highlighting the ongoing difficulty of verifying and fixing them.
Open questions and risks
The announcement leaves several practical questions unanswered. It is unclear how participating companies will safely test and deploy fixes without disrupting utility operations, which is one of the main challenges in securing critical infrastructure. Anthropic also did not specify whether partners will receive free access to models or who will cover the computing costs associated with running these AI tools.
Conclusion
Anthropic’s program aims to channel advanced AI and engineering expertise toward protecting critical infrastructure. Its ultimate effectiveness will depend on resolving operational issues around safe deployment, verification of findings, and the financial model for model access and computing resources.
