I work on the threat modeling1 team at GovAI, where I focus on cyber misuse risks from AI. My research examines extreme misuse scenarios to understand maximum potential harms and inform risk mitigation strategies. This work feeds into designing evaluations, setting capability thresholds for frontier safety policies, and prioritizing interventions. I focus on verifying that proposed safeguards are effective against realistic adversaries and that defensive tools are operationally viable for the defenders who need to use them.
Contact me at kamile [dot] lukosiute [at] governance [dot] ai.
Unfortunately, "threat modeling" here has nothing to do with cybersecurity threat modeling. What we mean by threat modeling is closer to the risk modeling definition in the EU AI Act Code of Practice. ↩