Preprint
Aug 2026
ADeptS-Bench: Measuring the Trustworthiness of Computer Use Agents Across Devices
ADeptS-Bench is introduced, a dual-stream trustworthiness benchmark, grounded in the ADEPTS capability framework and general population user studies, and reveals that all models overestimate consequence severity, mirroring the over-refusal bias observed in safety.
Joy Chen, Alejandro Castillejo Munoz, P. D'Oro et al.
· 0 citations