Teodora Baluta in Cybersecurity Dive: As frontier models go rogue, colleges beef up AI cybersecurity education

At the Georgia Institute of Technology, Assistant Professor Teodora Baluta is in the process of creating a permanent course out of a machine learning security elective she taught in the spring and fall of 2025. Her lessons connect social questions about AI, such as the privacy of training data, with technical considerations such as using threat modeling to prevent jailbreaking and preserve robustness.

The course’s goal is to teach students how to engage in threat modeling that compares a system’s defenses to the dangers it faces. Baluta said she wanted students to come away from the class “thinking a little bit adversarially.” Certain threats may not be very common, she added, “but if at least they’re aware of it conceptually, they’ll be more responsible AI computer scientists.”