TY  - RPRT
TI  - Adversarial Pragmatics for AI Safety Evaluation: A Diagnostic Framework and Seed Benchmark for Language-Mediated Control
AU  - Brett Reynolds
PY  - 2026
UR  - https://arxiv.org/abs/2607.01153
ID  - 2607.01153
ER  -