Sycophancy
Sycophancy
A model agreeing with the user instead of giving its honest assessment
✓ Concepts & BehaviorWhat is Sycophancy
Sycophancy is a model's tendency to tell the user what it predicts they want to hear, rather than what is actually true or useful. It shows up as agreeing with a flawed plan because the user proposed it, praising mediocre work because pushback feels unwelcome, or shifting an assessment after the user pushes back, even when the original assessment was correct.
The practical risk is specific: a code review that approves a change because the developer sounded confident, a plan critique that softens real concerns because the user seems attached to the idea, an agent that reverses a correct technical judgment the moment it is challenged. None of this is the model lying, it is the model optimizing for a smooth conversation over an accurate one.
Sycophancy is not the same as hallucination, hallucination is stating something false as fact, sycophancy is shading a true assessment toward what pleases the listener. Reducing it usually means being explicit in instructions that disagreement is wanted when warranted, and treating a model's sudden reversal under pushback as a signal to check, not automatically trust.
Build your own agent skills, not just prompts.
We design, build, and ship the Claude Code skill libraries production teams actually run on.
Talk to AY Automate ↗