OpenAI has announced the introduction of new security measures to its ChatGPT chatbot after cases of users turning to it during mental crises. The GPT-5 model received improved skills in recognizing risks of self-harm, redirecting people to helplines, and in critical situations, transmitting information to moderators or even law enforcement agencies.
“Our goal is to make our tools as useful as possible for people, and as part of this, we continue to improve how our models recognize and respond to signs of mental and emotional distress, as well as provide people with help guided by expert advice”, the company said.
The company cooperates with more than 90 doctors from 30 countries of the world, including psychiatrists, pediatricians and family doctors. In the US, the chatbot directs people to the 988 hotline, in the UK to Samaritans, and in other countries to findahelpline.com.
After the release of GPT-5, the accuracy of response to crisis situations increased by more than 25% compared to the previous version. A new approach to training, safe completions, allows the model to give useful but safe advice.
OpenAI is also working on enhanced protection in long dialogues and blocking of dangerous content. Among the new features are direct contact with emergency services, quick notification of trusted persons, as well as tools for teenagers and parents: parental controls and additional safeguards against risky content.
