epistemic guardrails

Why AI reading assistants need guardrails

A minimal prototype tests whether LLM reading assistants stay honest when users push them beyond retrieval into interpretation.