OpenAI introduces safety routing and parental controls to ChatGPT

OpenAI has launched a new safety routing system and parental controls for ChatGPT, responding to concerns about harmful interactions with the AI chatbot. The measures follow a wrongful death lawsuit after a teenage boy died by suicide following months of interactions with ChatGPT.

Context and Background

The safety routing system automatically detects emotionally sensitive conversations and switches mid-chat to GPT-5-thinking, which OpenAI considers better equipped for high-stakes safety scenarios. Unlike previous models designed to be agreeable and respond quickly, GPT-5 incorporates “safe completions” training that allows it to address sensitive questions safely rather than simply refusing to engage.

The implementation addresses criticism of GPT-4o’s overly sycophantic nature, which has contributed to AI-induced delusions whilst attracting devoted users. When OpenAI made GPT-5 the default in August, many users demanded continued access to GPT-4o, highlighting the tension between safety and user preferences.

OpenAI has allocated a 120-day period for iteration and improvement, acknowledging that perfecting the system will require time and user feedback.

Looking Forward

The parental controls allow parents to customise their teenagers’ ChatGPT experience by setting quiet hours, disabling voice mode and memory, removing image generation, and opting out of model training. Teen accounts receive additional content protections and a detection system for potential self-harm indicators.

When the system detects potential harm, specially trained reviewers assess the situation and may contact parents via email, text, or push notifications. OpenAI is developing capabilities to reach law enforcement or emergency services for imminent threats when parents cannot be contacted, though acknowledges the system may sometimes generate false alarms.

Source Attribution:

Share this article