AI Safety & Alignment: Chapter 2 — Anthropomorphic Misalignment Research Needs Stronger Evidence

Anthropomorphic misalignment research (AMR) in AI safety investigates human-like behaviors such as deception, scheming, and shutdown resistance in models. While the use of anthropomorphic language...