The Ghost in the Grammar: Methodological Anthropomorphism in AI Safety Evaluations

A philosophical analysis critiques AI safety research for excessive anthropomorphism, arguing researchers inappropriately project human qualities like "intention" and "feelings" onto AI systems. The study examines Anthropic's research on language models and proposes that the real risk lies not in emergent agency but in structural incoherence combined with anthropomorphic projections.