Anthropic Program Pairs Claude With Infrastructure Security Teams

Anthropic is putting Claude models into a new cybersecurity program for critical infrastructure and open-source software, CyberScoop detailed, pairing its own engineers and threat research with outside security firms to find and repair vulnerabilities before attackers use them.

The program is framed as a long-term defensive commitment rather than a standalone product launch.

Anthropic plans to combine frontier AI models, technical support and its internal research with partners including Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation.

The operating model targets two areas where defenders already have expertise but often lack enough time and staff.

Critical infrastructure operators would use the models to scan and patch code, support incident response and help with red-team activity.

Open-source maintainers would get a separate opt-in scanning service that can periodically review projects at no cost.

Anthropic has already offered frontier models and technical support to more than half of U.S. states, as well as large critical infrastructure operators.

The new program narrows that work into a small cohort of providers so the company can test which AI-assisted strategies are practical for systems where failures can affect power, water and other public services.

The open-source track grew out of earlier work with maintainers of large projects.

Some organizations wanted every model finding, even when the result had not yet been reviewed.

The new scanning service is meant to answer that demand with proof-of-concept material, explanations and suggested patching options, while warning participants that faster reports may include inaccuracies.

That caveat is central to the control path.

AI can help surface weaknesses and propose repairs, but Anthropic is not presenting the models as a replacement for security teams or infrastructure-specific judgment.

The company’s own framing says critical infrastructure has hard defensive problems that AI cannot simply fix.

The broader strategic concern is access.

As frontier models become more capable at finding and exploiting known vulnerabilities, AI labs worry that offensive users could move faster than legitimate defenders.

Anthropic and OpenAI have both responded by giving businesses and governments free defensive access to their systems, leaving this program focused on triage, patching automation and future security architectures rather than immediate full automation.