2 papers
cs.MA2026
Manipulation Is Task-Dependent: A Multi-Axis, Multi-Environment Evaluation of Frontier LLMs
Adeeb Zaman, Erik Nordby, Fred Heiding
We evaluate manipulative behavior in six frontier language models across six environments, ranging from negotiation tasks to agentic workflows, resulting in 13{,}590 individual sce…
cs.LG2026
Linear Probe Accuracy Scales with Model Size and Benefits from Multi-Layer Ensembling
Erik Nordby, Tasha Pais, Aviel Parrack
Linear probes can detect when language models produce outputs they "know" are wrong, a capability relevant to both deception and reward hacking. However, single-layer probes are fr…