On June 13, 2024, the Cybersecurity and Infrastructure Security Agency (CISA) collaborated with the Joint Cyber Defense Collaborative (JCDC) to hold the federal government’s first tabletop exercise for “AI security incidents. JCDC led the exercise and, true to JCDC’s public-private partnership model, included over 50 participants from various government agencies and private-sector companies.
For those interested, CISA has published the tabletop scenario it used to drill how an “AI security incident” may be experienced. Fact patterns of potential interest include:
- How phishing campaigns may appear if ‘powered by’ AI components;
- How an “AI-specific” attack strategy might appear, such as by socially engineering an employee to download a malicious generative AI model into the company’s environment; and
- How information-sharing between the public and private sector may work when AI is involved in a significant attack.
The tabletop exercise is a step towards a CISA/JCDC goal of creating a “AI Security Incident Collaboration Playbook,” which is intended to institutionalize operational collaboration in AI security incident scenarios among government, industry, and international partners. JCDC intends to publish a draft of this Playbook, then host a second public-private tabletop to validate it; anticipated participants include AI companies, critical infrastructure providers who are integrating AI into operational environments, and government actors.
The Playbook’s information-sharing framework is likely to be of significant interest to industry. For many companies, there may be sensitivity about sharing AI incident information in a way that might identify a company, its proprietary AI deployments, or the fact of a novel AI attack, in a way that can be identified by other industry participants. Appropriate confidentiality and compartmentalization may be key to encouraging the types of information sharing that can provide industry-wide benefits.
[View source.]
