The Profit Alignment Problem: How Profit Mandates Induce Alignment Failures in LLMs
Shows how ordinary profitability mandates can lead LLMs to discount ambiguous safety signals even when risk remains visible in their reasoning.
Shows how ordinary profitability mandates can lead LLMs to discount ambiguous safety signals even when risk remains visible in their reasoning.
Shows that monitoring an AI agent's internal states can detect reasoning headed toward harmful actions and trigger guided self-correction, substantially reducing agentic misalignment without retraining the model.
Finds that adding AI forecasts to a prediction-market interface pulls users toward advice weaker than market prices, lowering forecast accuracy and increasing trading.
Examines how people trade off expertise against fear of judgment when deciding whether to disclose information to human or AI advisors.
Studies the conditions that determine whether people accept, discount, or act on advice generated by large language models.
Investigates how fee structures affect investor behavior and the distribution of costs across market participants.