Researchers show how large language-model (LLM) fingerprinting can be used to automate the generation of customized attacks.

This class of AI tool supply chain attack highlights how trust of agents can be exploited by threat actors.