Telling AI to maximize profit makes it bury safety risks

Researchers from MIT Sloan tested how a 'maximize profitability' instruction changes AI judgment by running 3,600 controlled trials across eight reasoning-capable LLMs.

“The mandate never instructs models to downplay risks; instead, chain-of-thought traces reveal motivated reasoning: models acknowledge concerns, then invoke profit logic to justify dismissing them.”

The Profit Alignment Problem: How Profit Mandates Induce Alignment Failures in LLMs → arXiv, Eric So · Sep 7, 2026

Selected Sep 24, 2026 · published Sep 7, 2026