#artificial intelligence
Apr 2026
HINTBench: Horizon-agent Intrinsic Non-attack Trajectory Benchmark
These findings establish intrinsic risk auditing as an open challenge for agent safety, where intrinsic failures remain latent, propagate across long-horizon execution, and eventually lead to high-consequence outcomes.
Jiacheng Wang, Jinchang Hou, Fabian Wang et al.
· arXiv.org · 9 citations