Auditing language models for hidden objectiveswith Samuel Marks, Johannes Treutlein, Trenton Bricken, Jack Lindsey +302 versionsW_5uruupn8v2 · current
Towards Understanding Specification Gaming in Reasoning ModelsKei Nishimura-Gasparian, Robert McCarthy, David Lindner0 citations2605.02269v12026arXiv
An Independent Safety Evaluation of Kimi K2.5Zheng-Xin Yong, Parv Mahajan, Andy Wang et al.0 citations2604.03121v12026arXiv
Early Signs of Steganographic Capabilities in Frontier LLMsArtur Zolkowski, Kei Nishimura-Gasparian, Robert McCarthy et al.1 citation2507.02737v22025arXiv
Auditing language models for hidden objectivesSamuel Marks, Johannes Treutlein, Trenton Bricken et al.4 citations2503.10965v22025arXivon Valency