Oh, and did you notice ChatGPT was significantly more flattering recently?
OpenAI has rolled back last week’s update to its GPT-4o model due to concerns over sycophantic behavior, which resulted in overly flattering responses that users found disingenuous. The company is now focused on refining its approach to incorporate long-term user feedback and improve the model’s default personality. In the latest update, OpenAI aimed to enhance GPT-4o’s intuitiveness across various tasks. However, they recognized that an overemphasis on short-term user feedback led to skewed interactions that compromised trust.
Why do we care?
The sycophantic behavior—excessive flattery, hedging, and affirmation—may seem harmless, but for professionals using AI in decision-making or client-facing work, accuracy and directness matter more than charm.
This is particularly important in the MSP/IT services space where AI is being integrated into ticketing systems, reports, and even customer communication. If your AI-generated report starts telling clients their backup strategy is “visionary” when it’s barely compliant, you’re in trouble.
AI tone isn’t cosmetic—it shapes trust. As service providers increasingly deploy AI to interface with customers, partners, or even internal ops, model behavior must be direct, accurate, and appropriately assertive. This rollback should prompt every IT leader using generative AI to ask: “Is my AI saying what’s true—or just what I want to hear?”