OpenAI moves to curb ChatGPT sycophancy

OpenAI is making ChatGPT less automatically agreeable, aiming to have it flag errors and offer alternative views to counter "sycophancy." Its own data show GPT-5's agreeable responses fell from 14.5% in GPT-4o to under 6%, a 69% drop for free users and 75% for paid subscribers.

• Stack Overflow for Agents triples signups with ChatGPT plugin2026-10-16
• States sue OpenAI, saying ChatGPT harms children and deceives adults2026-10-16
• ChatGPT's Chrome extension can now read your open tabs2026-10-16
• ChatGPT Ads Target Users Based on Prompts and Contact Info2026-10-15
• Arkansas AG Sues OpenAI Over ChatGPT Safety2026-10-14