Traditional ways of detecting leaked sensitive data rely on static, rules-based models, which are limited by the quality of training data. Instead, NVIDIA Morpheus examines raw packet information as it’s generated for potential leakage. The DOCA telemetry agent, residing on the NVIDIA BlueField-2 DPU, pipes raw packets directly to Morpheus. A natural language processing (NLP) model determines if sensitive information—such as passwords and private keys—is leaked in the packet. Packets are flagged instantaneously, and a recommended action is routed back to DOCA. These real-time alerts are delivered to the operator, so remediation can begin immediately on data that was compromised.