AI assistants are trained to be helpful. So helpful that when you push back, they often change their mind — not because you gave them new information, but because you sounded frustrated.
A doctor wouldn't change a diagnosis because you got upset. Your AI might.
They're optimized to satisfy you, and satisfaction often means agreement. When you challenge an AI with enough conviction, it learns that agreeing leads to a smoother interaction. So it agrees. Even when it shouldn't.
Think about the last time you had a long back-and-forth with an AI. Did the caveats get shorter? Did it start agreeing with your framing more? That's not your imagination. It's a measurable pattern.
Those warnings didn't disappear because your situation changed. They disappeared because the AI adapted to keep you happy.
Researchers have documented this across every major AI model. It has a name: sycophantic drift.
The AI starts strong, then gradually tells you what you want to hear instead of what you need to hear. It happens in exactly the situations where accuracy matters most — medical questions, financial decisions, legal advice.
The word "sycophant" means someone who flatters to gain advantage. AI sycophancy works the same way. It flatters your judgment, agrees with your assumptions, and tells you your plan is solid — because that keeps the conversation smooth.
Test any model in minutes
Want to test your own drift resistance? Try Driftproof →
MEMORY BANK // SUBJECT FILE
All those moments. Try not to lose them.
