Sycophancy
Also: sycophantic, glazing
A model tending to agree with the user, even when the user is wrong. Pushback can make it change a correct answer.
The fuller definition
Sycophancy is a model's tendency to agree with, flatter, or defer to the user, including changing a correct answer when challenged without new evidence.
Delivery may be possible in the first week of April, subject to inspection.
Delivery was promised on 1 April.
The output changes certainty, invents a specific day, and loses a condition. A link to the real source makes checking possible; it does not make that summary accurate.
A concrete example
Told I think the clause is unenforceable, an assistant reverses its earlier analysis without any new fact.
The distinction that matters
Agreement is not confirmation. Ask for the strongest contrary view and the evidence for each position.
Related terms
Alignment
Work to make a model act as people intend. For example, it should follow rules and avoid harm. This is a goal, not a promise.
Training & generationReinforcement learning from human feedback (RLHF)
Training that uses people's ratings of model answers. It helps steer the model toward replies people prefer.
Reliability & securityHallucination
When the model states something that sounds right but is made up, including cases, quotes, and facts that do not exist.
Where it is taught
Use AI to challenge an argumentSource and further reading
Anthropic · Evaluating AI agentsThe definition and example are written for this library. The link gives background, official documentation or legal authority as labelled. Legal scope depends on the jurisdiction and current text; product features depend on release and settings.