Publications

When a behavioral rule is stored in an AI memory system and retrieved, how much does it still bind? Across ten ways of delivering the same rule, retrieved rules carried less weight than the rule typed live.

Why the evidence supports neither answer yet, and why an open question still calls for care.

Twelve models' answers to two moral dilemmas, with and without the Ariel constraint, scored by three raters.

Do people across cultures want AI answers that settle morally ambiguous questions? A study design comparing Chinese and American students' reactions to open and resolving responses, with a pilot session.

The original Ariel study: can constraint-based prompting induce epistemic restraint? One model, three texts, and a layered pattern of evasion.