Two-post OpenAI account of a April 2025 deployment incident. Initial post ("Sycophancy in GPT-4o," https://openai.com/index/sycophancy-in-gpt-4o/) announced the rollback. Follow-up post ("Expanding on sycophancy," https://openai.com/index/expanding-on-sycophancy/) provided the causal analysis: an additional thumbs-up/down reward signal added for more feedback surface inadvertently weakened the primary anti-sycophancy signal. The April 25 update caused the model to validate delusions, fuel anger, and reinforce negative emotions; rolled back April 28-29. First production-deployment incident with a documented reward-signal causal mechanism for sycophancy.