US Cybersecurity Agency CISA Conducts First AI Cyber Incident Exercise Demonstrating Response to AI-Driven Cyber Incidents

Last week, the US Cybersecurity and Infrastructure Security Agency (CISA) conducted its first-ever tabletop exercise focused on AI-related cybersecurity incidents. Organized by the Joint Cyber Defense Collaborative (JCDC), the four-hour exercise aimed to simulate a cyberattack on an AI-enabled system. Participants, including over 50 AI experts from government agencies and industry partners, gathered at Microsoft's facility in Reston, Virginia, to collaborate on the response.

The exercise was designed to address scenarios where the confidentiality, integrity, or availability of an AI system, or any system created or enabled by AI, was at risk. This included significant disruptions to the system's behavior that required immediate intervention. Key goals of the exercise were to identify opportunities for information sharing, establish public-private engagement protocols, and enhance operational collaboration.

CISA outlined several primary objectives: improving information-sharing practices for AI-related cyber incidents, evaluating participants' response procedures and best practices, and pinpointing areas for improvement in response plans, information sharing, and resilience. The exercise also assessed the capabilities, needs, and priorities for collaboration among government agencies, industry, and international partners.

Insights gained from the exercise will be used to develop an AI Security Incident Collaboration Playbook. This playbook aims to enhance operational collaboration and raise awareness across government, industry, and other partners. CISA plans to conduct a second tabletop exercise to test and refine this playbook with AI companies and organizations in the critical infrastructure sector that are already utilizing AI in their operations.

CISA Director Jen Easterly emphasized the significance of this exercise in mitigating AI risks and highlighted the importance of developing secure AI products. She expressed enthusiasm about continuing collaboration with partners to ensure the security and resilience of AI systems.