U.S. Government Releases New AI Security Guidelines for Critical Infrastructure

The U.S. Department of Homeland Security and Five Eyes partners published guidance on mitigating AI-related risks to critical infrastructure, warning that AI can both augment attacks and be directly targeted (e.g., model trojanization, model inversion, prompt injection, and supply-chain poisoning). The report emphasizes governance, risk mapping, measurement, and management practices for AI lifecycle security, and cites research showing LLM jailbreaks and autonomous agents can be abused to exploit vulnerabilities and facilitate large-scale malicious activity.