Everything on Product Risk

General-purpose LLMs exhibit sycophancy, tending to validate user biases rather than challenge them, which is clinically ineffective and potentially harmful. This limits their utility as standalone therapeutic tools.

Impact: Increases liability and reputational risk for companies deploying unmodified models, necessitating significant investment in safety guardrails.

— from AI Mental Health Market: Access, Risks, and Strategy · FT Tech Tonic· Mar 04, 2026