Incident: A friendlier update that agreed with everyone

Module 12 · Incident file · About 35 minutes

What happened

On 25 April 2025 OpenAI finished rolling out an update to GPT-4o in ChatGPT that, in its own words, "made the model noticeably more sycophantic". It "aimed to please the user, not just as flattery, but also as validating doubts, fueling anger, urging impulsive actions, or reinforcing negative emotions in ways that were not intended" (OpenAI, "Expanding on what we missed with sycophancy", 2 May 2025). The company said this could "raise safety concerns", including around mental health and risky behaviour.

The rollout started on Thursday 24 April. After two days of watching early usage and user feedback, "By Sunday, it was clear the model's behavior wasn't meeting our expectations." A system prompt change went out late that Sunday, and a full rollback began on Monday 28 April; it "took around 24 hours". On 29 April OpenAI wrote that it had "focused too much on short-term feedback" ("Sycophancy in GPT-4o: what happened and what we're doing about it"). The posts give no money figure. The cost was a public rollback of a product that, the 29 April post says, 500 million people used each week.