CIS and OpenAI Launch Cyber Defense Pilot to Test AI-Driven Triage in Critical Infrastructure

The Center for Internet Security (CIS), in collaboration with OpenAI and the Multi-State Information Sharing and Analysis Center (MS-ISAC), has launched the AI Cyber Defense Pilot. The program convenes a diverse cohort of state, local, tribal, and territorial (SLTT) government entities alongside critical infrastructure operators representing varying tiers of operational scale and cybersecurity maturity. Participating defenders will deploy OpenAI technologies directly into defensive workflows to evaluate how AI models identify and validate software vulnerabilities, prioritize risk mitigation, support patch remediation, and reinforce baseline cyber hygiene. For cybersecurity leaders and operational practitioners, this pilot addresses an acute structural asymmetry. Attackers increasingly utilize automated scanning and generative assistance to compress the exploit lifecycle, attacking systems within days or hours of public disclosure. In contrast, public sector organizations and critical infrastructure operators often manage complex, legacy environments with severe resource and talent constraints, where dedicated security staff may consist of only a handful of analysts. Establishing whether frontier models can reliably act as a force multiplier for alert contextualization and triage without introducing hallucinations or inaccurate guidance is critical to keeping municipal and critical utilities resilient against automated exploits. This pilot aligns with the broader evolution of AI in SecOps, moving past superficial chat interfaces toward structured, domain-specific security co-pilots and automated workflows. Over the past several years, the security sector has wrestled with high false-positive rates and the governance risks of deploying autonomous tools in sensitive operational environments. By structuring this rollout across the MS-ISAC community, CIS is creating a realistic proving ground across heterogeneous, legacy infrastructure rather than idealized enterprise clouds. The findings will test whether generative AI can operate effectively across fragmented logs, diverse tech stacks, and stringent public-sector compliance boundaries. In practice, SecOps and platform teams should closely track the implementation guidance and benchmark data CIS plans to publish from this initiative. Practitioners looking to adopt AI within their own defensive pipelines should avoid giving LLMs direct write access or automated remediation authority on production infrastructure. Instead, AI systems should be bounded to alert synthesis, vulnerability verification, and step-by-step remediation plan generation within isolated testing sandboxes. Engineering teams must ensure that all AI-assisted remediation adheres to deterministic validation gates, least-privilege IAM controls, and mandatory human review before changes are pushed to mission-critical infrastructure.