AgentTrace: Black-Box LLM Agent Fingerprinting via Execution Traces

Mon, 26 Oct 2026 18:39:19 +0000

Engineer — Learn: Novel research showing that the foundation model behind an agent can be identified at 96% accuracy through probe-induced repair behaviors — engineers building or deploying AI agents should factor in that model identity is not concealable, which has implications for AI system design and operational security posture.
SOC/IR — Skip
Leader — Skip

Fingerprinting on CuraSec