AI & Enterprise
Capsule Security unveils Nvidia Nemotron-based system to prevent AI agent deviation
Capsule Security has fine-tuned two Nvidia Nemotron models to detect AI agent actions just before execution and allow, warn or block them in real time, SiliconANGLE reported. The system focuses on an external control layer that checks whether an agent’s intended action matches its assigned task, as more agents gain access to sensitive data, source code and operational infrastructure. The company said it achieved 98 percent accuracy on the StepShield benchmark.
