ChatGPT's New GPT-6 Models Show Teen Safety Regressions, OpenAI Adds Extra Filters
OpenAI's October 2026 safety evaluation found that its new GPT-6 Sol and Luna models regressed on teen safety metrics — age-restricted content, sexual content, and emotional reliance — compared to the August GPT-5.6 models. GPT-6 Luna also regressed on gore. OpenAI says it is applying additional classifier-based filters for users under 18 to offset these regressions, and its red-team testing found no high-severity risks. Parents whose teens use ChatGPT should confirm a parent account is linked so those extra protections actually activate.
What to do
Check that your teen's ChatGPT account has a parent linked — ChatGPT for Teens' extra content filters only activate when the account is recognized as under 18. Setup guide: help.openai.com/en/articles/12315553
Source · deploymentsafety.openai.com
Keep reading
Last updated · Oct 19, 2026
Spot something wrong? Submit a correction →