Latent Fact-Checking: Detecting Misinformation through Activation Engineering

A collection of new arXiv computer science and machine learning papers covering misinformation detection with activation engineering, risk-aware agent policies, LLM oversight and adversarial robustness, causal intervention falsification, operator learning, clinical EHR modeling, SmartNIC-accelerated graph neural network training, crystal structure prediction, and model interpretability. The material is primarily academic and does not describe a specific cybersecurity vulnerability or exploit.