An October 2026 arXiv preprint introduces DANGERLENS to examine whether models with matching refusal behavior can still differ in internal accessibility of dangerous concepts…
A new arXiv preprint shows GPT-5.4 agents produce varying interpretations of the same scientific claim across runs, even when using structured Claim Cards. The research…
arXiv preprint tests a disagreement-driven method that rewrites only contested rubric clauses without human labels. It increases inter-judge reliability on one dataset yet…
New arXiv research traces weak refusal circuits already present in pretrained models and shows that confining safety updates to these circuits cuts harmfulness by 63 percent…
arXiv preprint finds that safety alignment in Looped Language Models can weaken at deeper recurrent steps under jailbreak attacks, with cross-depth transfer observed. Standard…
An arXiv preprint released October 7 2026 introduces GUISE, a cross-language benchmark measuring how narrative wrappers dramatically raise attack success rates on…
A new arXiv preprint introduces SIGMA, a pipeline where models use only a written Model Spec to synthesize alignment dilemmas, then train themselves via supervised fine-tuning…
An arXiv preprint dated October 2 2026 applies Many-Facet Rasch Models to summarization ratings and finds that moderate alignment in latent summary quality does not guarantee…
The UK AI Security Institute is publishing verified benchmark results for frontier models using EvalEval's Every Eval Ever schema and Evaluation Cards, advancing reproducible…
A July research preprint examines changes in decision behavior after abliteration. Its narrow experiment raises a wider question about model modifications.