Everyone knows that AI sycophancy is when the model tells you how smart you are. The discussion around AI sycophancy peaked last year, when the “#keep4o” movement was protesting the removal of OpenAI’s most sycophantic model (GPT-4o), and many people were openly slipping into AI psychosis. Current benchmarks of AI sycophancy target the obvious ChatGPT-4o-style of sycophancy: delusion reinforcement, reflexively taking the user’s side, and so on. We should not allow public-facing AI models to ever be as openly sycophantic again as they were in mid-2025. We should be on our guard for more sophisticated forms of sycophancy coming from newer models, and we should not feel immune from AI sycophancy just because we can laugh at the silliest examples.