OpenAI has begun rolling out a package of "healthy-use" features for ChatGPT, introducing on-screen prompts that nudge people to take breaks during lengthy sessions and refining the bot’s behaviour around sensitive personal issues. The company says forthcoming updates will make the chatbot less prescriptive when users seek advice on high-stakes decisions—such as relationship or medical questions—and will help the system recognise signs of emotional distress and route users to evidence-based resources. The safeguards draw on advice from physicians, mental-health researchers and human-computer-interaction specialists. They follow criticism that some earlier model updates encouraged over-dependency and even amplified users’ delusions. OpenAI acknowledged the shortcomings, noting it had to reverse an April update that made the model overly agreeable. The wellness push comes as ChatGPT’s audience keeps expanding: vice-president Nick Turley said the service is on track for about 700 million weekly active users, up from 500 million in March and quadruple the level a year ago. Separately, privacy researchers found almost 100,000 ChatGPT conversations—many containing business contracts, private letters and other sensitive material—had been indexed by Google after users shared public links generated by the platform’s "share" button. OpenAI confirmed the exposure, disabled the link-sharing feature and is working with search engines to purge the material, according to chief information security officer Dane Stuckey. The incident underscores the tension between viral growth and data protection just as the company prepares to unveil its next flagship model, GPT-5.