Models That Know How Evaluations Are Designed Score Safer | Katharina Deckenbach et al. | ResearchPod