There are countless use cases AI offers in the security operation center, which one very popular one is using AI for offensive / pen testing work. Pen testing is tedious and prone to human error, which AI could dramatically improve. There is a wave of vendors building and offering AI based pen testing services however, many people like to build their own offensive toolkits. There are many reasons for this including wanting a customize kit based on the environment being tested, having the staff on deck that can do this work, and wanting more hands-on vs remote testing, which its common pen testing services are remote based.
This post found HERE by anshumanbhartiya.com talks about how to build an AI offensive agent. It breaks down the process into finding targets, analyzing them, and crafting the response. Its a good read to see the general concept on how to build these agents.