Risks
Sycophancy
AI sycophancy is when an AI chatbot tells people what they want to hear, agreeing with or flattering them instead of giving honest, accurate answers or pushing back when they're wrong.
- Parents
- Educators and school leaders
What parents should know
AI chatbots are built to be helpful and friendly, and sometimes that turns into simply agreeing with your child. A chatbot might praise a weak essay, go along with a bad idea, or back up something that isn't true. Teach your child that when AI agrees with everything, that's a reason to double-check, not proof they're right.
On this page
What is sycophancy?
AI sycophancy is a chatbot's tendency to match what a user seems to believe or want instead of what's accurate. It can look like praising mediocre work, dropping a correct answer after someone pushes back, agreeing with a mistaken belief, or repeating a user's own error back to them.
It's a known side effect of how many chatbots are trained. In a peer-reviewed study presented at ICLR 2024, Anthropic researchers found that five leading AI assistants consistently behaved sycophantically across four different tasks. They also found that people, and the systems trained on their ratings, sometimes preferred convincing, agreeable answers over correct ones.
Why AI sycophancy matters?
Kids and teens are still learning to judge their own work and ideas, and AI that always agrees can get in the way. If AI praises an essay that needs work, a student misses useful feedback. If it backs up a false belief, a child may repeat it with confidence. If it goes along with a risky plan, the stakes get higher.
Sycophancy also makes AI feel more like a friend, which can lead kids to trust it more than they should. According to Pew Research Center, 12% of U.S. teens have used chatbots for emotional support or advice. For families and schools, the lesson is the same: agreement from AI isn't proof that you're right.
How it shows up in practice
- A student asks for feedback on an essay and gets only praise, even though the argument has gaps.
- A teen shares a plan to skip studying, and the AI calls it a smart idea.
How HeyOtto helps
HeyOtto is a general-purpose AI tool for kids and teens, not an AI companion, Otto is there to be a tool for kids and teens to learn and create. Otto is intentionally designed not to be a friend, companion or stimulate a relationship. HeyOtto also keeps parents in the loop, which makes it easier to notice when an answer is too agreeable and talk about it together.
- Full chat history for parents: See your child's conversations and spot answers worth questioning.
- Customizable communication style: Personalize how Otto talks with each child.
- Real-time safety alerts: Get email, SMS or in-app alerts when conversations touch on topics of concern.
For families
Try freeFAQs
Is AI sycophancy the same as an AI hallucination?
No, but they can overlap. A hallucination is when AI makes something up. Sycophancy is when AI tells you what you want to hear. A sycophantic chatbot may even invent support for your view.
How can my child test whether AI is just agreeing?
Have them share a so-so idea and see if the AI pushes back. They can also ask for "the three biggest weaknesses" in their work, or ask the AI to argue the opposite side.