Anthropic Puts Frontier AI to Work Defending Critical Infrastructure

Anthropic launched the Anthropic Cyber Mission on October 8, 2026, to help defend systems that communities depend on. The initiative will provide frontier Claude models, engineering support and threat research to critical-infrastructure providers and open-source maintainers.

The program focuses on operational technology behind power grids, water systems, factories and transportation networks. Anthropic says these systems can be difficult to patch because they often cannot be taken offline safely, allowing known vulnerabilities to remain unresolved for years.

What Anthropic is launching

The Critical Infrastructure Defense Program brings Claude models, on-site engineers and Anthropic threat research to trusted providers that secure operational technology. Its founding partners are Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation.

Anthropic is also launching OSS Scanner, an opt-in service that periodically scans participating open-source projects for security problems at no charge. Reports can include a proof of concept, an explanation of the vulnerability and a suggested fix.

The reports are generated by models without human review. Anthropic expects a true-positive rate above 90%, but says findings can contain errors, including incorrect severity ratings. Projects that enroll must therefore validate findings before taking action.

Anthropic says the work is already underway with a small cohort of providers. It also says it has offered frontier Claude models and technical support to more than half of U.S. states and some large public critical-infrastructure operators through a related cyber-defense program launched in June.

Why the timing matters

Recent reporting has described a widening gap between the speed of automated attacks and the ability of organizations to patch aging systems. The Atlantic reported that criminals and state-sponsored groups are using AI to conduct cyber operations faster and at greater scale, while hospitals, utilities and other essential-service providers often have limited resources and complex systems.

Anthropic’s approach combines AI assistance with established security vendors and infrastructure specialists rather than presenting a chatbot as a replacement for them. The company also acknowledges that AI cannot by itself solve problems involving outdated equipment, limited budgets or the risks of changing software connected to physical systems.

How open-source maintainers can get started

1. Review Anthropic’s Cyber Mission announcement to determine whether the project is suitable for OSS Scanner.
2. Use the OSS Scanner resources to review enrollment information for an eligible project.
3. Validate every model-generated finding, reproduce issues safely and coordinate fixes through the project’s normal disclosure process.

Anthropic says it plans to expand the program to more infrastructure sectors, open-source projects and security partners. Its larger test will be whether faster vulnerability discovery leads to verified patches and safer systems without introducing new risks.

References

Anthropic. (2026, October 8). Introducing the Anthropic Cyber Mission. https://www.anthropic.com/news/anthropic-cyber-mission

Wong, M. (2026, October 7). How long until AI hacks everything? The Atlantic. https://www.theatlantic.com/technology/2026/10/ai-hacking-cybersecurity-race/688911/