Reva: Real-Time Vulnerability Detection through Contextual Probing of Large Language Models

Summary

REVA is a real-time vulnerability detector that places a pre-computed threat model, lightweight static analysis findings and call-graph information in an LLM's context, then reads the model's internal representations with a trained probe in a single forward pass. On PrimeVul's held-out test set, REVA on GLM-5.3-Flash reaches 60.0% accuracy at about one second per function, matching the agentic detector VulTriage (59.2%) while running over 250 times faster. The probe also outperforms the LLM's own direct verdict by 14% to 17%.