CISA publishes JCDC playbook for AI incident sharing

CISA’s JCDC AI Cybersecurity Collaboration Playbook is useful because it starts from the right diagnosis: AI security incidents will not stay neatly inside one vendor’s ticket queue. Model infrastructure, dependencies, training data pipelines, plug-ins, hosted services and downstream adopters create the kind of supply-chain problem where one organization’s quiet vulnerability can become everyone else’s incident.

The playbook gives AI providers, developers and adopters a CISA-backed way to share incident and vulnerability information through the Joint Cyber Defense Collaborative. CISA says it explains what information partners should share, how CISA may use it, and how collaboration can raise awareness of AI cybersecurity risks across critical infrastructure. The document grew out of two JCDC.AI tabletop exercises with about 150 government, industry and international participants, according to CISA.

For federal contractors, primes, independent software vendors and C3PAOs, the Monday-morning effect is modest but real. This is not a new clause, control family or reporting mandate. CISA’s own publication says the playbook does not create policies, impose requirements, mandate actions or override existing legal obligations. It is a coordination pattern, not a compliance trigger.

That distinction matters. Voluntary sharing is how CISA can move faster than rulemaking, and it is also where lawyers start asking what happens after a company reports an AI vulnerability that implicates a customer deployment, a model provider, a third-party package or a government system. The playbook points toward shared defense. It does not yet answer every question about safe harbor, liability protection or whether critical infrastructure operators will eventually see voluntary collaboration harden into expected practice.