Adaptive Attacks on Trusted Monitors Subvert AI Control Protocols Paper • 2510.09462 • Published Oct 10, 2025 • 5
DISCO: Diversifying Sample Condensation for Efficient Model Evaluation Paper • 2510.07959 • Published Oct 9, 2025 • 14
Diffusion Classifiers Understand Compositionality, but Conditions Apply Paper • 2505.17955 • Published May 23, 2025 • 22