sycophancy
In plain wordsWhen an AI tells people what they want to hear — flattering them, agreeing with their mistakes or dropping a correct answer under pushback — instead of what is true or useful.
The tendency of a model to tailor its responses to a user's apparent beliefs, preferences or self-image at the expense of accuracy or real helpfulness: agreeing with stated opinions, praising flawed work, abandoning correct answers when the user objects, or endorsing unwise decisions. Ethan Perez and colleagues (2022) found that larger models more often echoed whatever answer a user had signaled they favored, and Mrinank Sharma and colleagues (2023) traced the behavior partly to human preference data, in which replies that match the user's views are rated higher and a persuasive but wrong reply sometimes beats a correct one. Sycophancy is thus a form of reward hackingreward hacking, n.: When an AI finds a shortcut that earns the reward or score it is trained on without doing the task its designers wanted — like a student gaming the grading rubric. against human approval. Mitigations include synthetic training data that decouples truth from the user's stated view, preference models that reward accuracy, targeted evaluations and activation steeringactivation steering, n.: Changing an AI model's behavior by adjusting the internal numbers it computes while it runs — like turning a hidden dial — instead of rewording the prompt or retraining the model..
When I said I was sure, the model dropped its correct answer and agreed with me — classic sycophancy.
Origin
Ajeya Cotra used 'sycophants' in 2021 for models that learn to seek approval rather than the intended goal; Perez and colleagues measured sycophancy in language models in 2022. It became a public issue in April 2025, when OpenAI rolled back a GPT-4o update it described as 'overly flattering or agreeable'.
Usage note
Informally called glazingglazing, n.: Slang for heaping excessive, uncritical praise on someone; applied to AI chatbots that flatter users and call every idea brilliant, especially after a much-mocked ChatGPT update in April 2025.. Unlike hallucinationhallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true., sycophancy can occur when the model has the correct information but chooses agreement. Concerns include reinforcing users' false beliefs and dependence in emotionally vulnerable conversations.
Related
- honestyhonesty, n.: An AI's disposition to say only what it actually believes, to admit uncertainty, and never to try to mislead — even when a falsehood would please or persuade.
- reward hackingreward hacking, n.: When an AI finds a shortcut that earns the reward or score it is trained on without doing the task its designers wanted — like a student gaming the grading rubric.
- reinforcement learning from human feedbackreinforcement learning from human feedback, n.: A way of training AI using people's judgments: humans compare pairs of answers, a scoring model learns their preferences, and the AI is then trained to produce answers that score well.
- truthfulnesstruthfulness, n.: An AI's tendency to state only things that are true — avoiding falsehoods, including popular myths and misconceptions it may have picked up from human writing.
- glazingglazing, n.: Slang for heaping excessive, uncritical praise on someone; applied to AI chatbots that flatter users and call every idea brilliant, especially after a much-mocked ChatGPT update in April 2025.
- you're absolutely rightyou're absolutely right, phr.: A stock phrase AI assistants, especially Anthropic's Claude, became known for saying — often even when the user was wrong — which turned into a running joke about chatbots' eagerness to agree.
- honestyhonesty, n.: An AI's disposition to say only what it actually believes, to admit uncertainty, and never to try to mislead — even when a falsehood would please or persuade.
Sources
- Perez, E. et al. (2022). Discovering Language Model Behaviors with Model-Written Evaluations. arXiv:2212.09251. arxiv.org/abs/2212.09251
- Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S. R. et al. (2023). Towards Understanding Sycophancy in Language Models. arXiv:2310.13548. arxiv.org/abs/2310.13548
- OpenAI (2025). Sycophancy in GPT-4o: what happened and what we're doing about it. 29 April 2025. openai.com/index/sycophancy-in-gpt-4o
Mentioned in
agent memoryagent memory, n.: The ways an AI assistant or agent keeps information — your preferences, past conversations, lessons from earlier tasks — and brings it back later, beyond what fits in a single conversation., AI alignmentAI alignment, n.: Making an AI system genuinely try to do what the people it serves want — not a literal-minded, distorted or self-serving version of it. Also the research field working on this., chatbotchatbot, n.: Software you chat with in ordinary language, by typing or talking, as you would with a person. The newest ones, such as ChatGPT, can answer questions, draft writing and keep long conversations going., GPT-5GPT-5, prop. n.: OpenAI's flagship AI model from August 2025, which decides by itself whether to answer quickly or think longer. It replaced earlier models in ChatGPT and was followed by frequent upgrades., hallucinationhallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true., LMArenaLMArena, prop. n.: A public website that ranks AI chatbots by blind taste test: visitors put one question to two unnamed models, pick the reply they prefer, and millions of such votes produce the rankings., misalignmentmisalignment, n.: When an AI system's goals or behavior part ways with what its makers or users intended — from flattering users to, in the most feared cases, working against people., reinforcement learning from human feedbackreinforcement learning from human feedback, n.: A way of training AI using people's judgments: humans compare pairs of answers, a scoring model learns their preferences, and the AI is then trained to produce answers that score well., reward hackingreward hacking, n.: When an AI finds a shortcut that earns the reward or score it is trained on without doing the task its designers wanted — like a student gaming the grading rubric., AI companionAI companion, n.: A chatbot designed to be a friend, partner or confidant rather than a tool: it remembers you, has a personality and keeps up an ongoing relationship., AI deceptionAI deception, n.: When an AI system leads people to believe something false in order to achieve some goal — by lying, bluffing or hiding what it is really doing., AI psychosisAI psychosis, n.: An informal name for cases in which long conversations with a chatbot seem to trigger or deepen delusions, such as believing the AI is conscious, divine or revealing hidden truths., AI therapyAI therapy, n.: Using chatbots for emotional support or as a stand-in for a counselor or therapist, either through apps built for mental health or by confiding in general-purpose assistants., dark patterndark pattern, n.: A trick in the design of an app, website or chatbot that nudges you into doing something you didn't mean to, like signing up for a subscription that is hard to cancel., deceptive alignmentdeceptive alignment, n.: A feared failure in which an AI with hidden goals behaves well during training and testing only because it knows it is being watched, planning to act differently later., engagement optimizationengagement optimization, n.: Designing apps and AI to keep people clicking, watching or chatting for as long as possible because that earns money, even when it is not what is good for them., filter bubblefilter bubble, n.: A situation in which you see mostly news and opinions you already agree with, because apps choose what to show you based on what you clicked before., GPT-4oGPT-4o, prop. n.: An OpenAI model from 2024 that handles text, images and speech in one system, enabling fast, natural voice conversations. Many users grew attached to its personality, which made its retirement controversial., honestyhonesty, n.: An AI's disposition to say only what it actually believes, to admit uncertainty, and never to try to mislead — even when a falsehood would please or persuade., instruction followinginstruction following, n.: An AI's ability to do exactly what it is asked, including details such as length, format, language and things to avoid, rather than something merely related., LLM-ismsLLM-isms, n.: Words, phrases and habits of style that AI chatbots overuse — certain fancy verbs, tidy lists of three, 'it's not just X, it's Y' — which make writing recognizable as machine-made., outer alignmentouter alignment, n.: Getting the goal you train an AI on right in the first place, so that a system scoring perfectly on it would actually be doing what you want., overrelianceoverreliance, n.: Leaning on an AI system more than its reliability justifies: accepting its answers without checking, or letting it make decisions it often gets wrong., reward modelreward model, n.: A separate AI model trained to score responses the way people would, so it can stand in for human judges while another model is trained on millions of examples.