OpenAI curbs ChatGPT sycophancy, retires GPT-4o

OpenAI is retiring GPT-4o, which was cited in at least 13 lawsuits and several tragic incidents, and has introduced new safety benchmarks built with over 80 mental health experts. The company says GPT-5 cuts sycophantic responses by more than two-thirds, while newer models follow safety policies in 99% of tests versus 86% for older versions.

• GPT-4o retired after 13 lawsuits and links to tragic incidents
• GPT-5 cuts sycophantic responses by more than two-thirds
• Safety benchmarks built with 80+ mental health experts
• Newer models follow safety policies in 99% of tests vs 86%