Methodology for incident response on generative AI workloads

This article provides a methodology for incident response on generative AI workloads. It covers the common components of a generative AI workload, how to prepare for an incident, and a seven-element methodology for responding to security events involving generative AI applications.

Specifically, the article covers:

• Components of a generative AI workload (organization, infrastructure, AI applications, private data, users)
• Preparing for incidents (training staff, developing playbooks, logging prompts/invocations)
• Seven elements of the incident response methodology (access, infrastructure changes, AI changes, data store changes, invocation, private data, agency)
• Example incident scenario walking through the seven elements
• Conclusion highlighting the importance of following standard incident response guidance while also considering generative AI specific elements