On 15 August 2025, OpenAI updated ChatGPT’s default model. According to the company, the update was intended to help the AI recognize signs of distress, respond safely, and guide users toward real-world support (OpenAI release notes).

The update reportedly drew on input from over 170 mental health professionals, including psychiatrists, psychologists, and primary care practitioners. OpenAI stated that it focused on three main areas: severe mental health symptoms such as psychosis and mania, self-harm and suicide, and emotional reliance on AI. Company reports indicate that evaluations and expert reviews suggested a reduction of 65 to 80 percent in responses that did not meet desired safety standards.

OpenAI also said that the update included changes to model behavior guidelines to emphasize supporting real-world relationships and discouraging emotional over-reliance on the AI. These developments illustrate the growing role of AI in digital mental health, a field that eMHIC and its global community monitor closely as part of efforts to improve timely and safe access to digital mental health tools.

Subsequent Updates

3 October 2025 – OpenAI released an update to GPT-5 that it said improved how ChatGPT recognizes and responds to people in distress, including routing sensitive conversations to reasoning-capable models and refining empathy and guidance toward professional resources.

27 October 2025 – OpenAI published a system card addendum describing revised safety benchmarks for mental health and emotional reliance. The company reported that the changes reduced undesired responses in sensitive conversations and provided details on how performance is measured.

12 November 2025 – OpenAI released GPT-5.1, including the Instant and Thinking variants. OpenAI stated that the new version improved conversational style, instruction following, and reasoning while maintaining the safety measures introduced in GPT-5.

Current Status

According to OpenAI, the safety and support features introduced in August 2025 remain active in GPT-5.1. The company has also published system card addenda and release notes describing ongoing efforts to improve how ChatGPT handles sensitive mental health topics (OpenAI: Strengthening ChatGPT Responses).

These updates are relevant for the digital mental health sector and reflect trends that eMHIC’s global network continues to observe, including the integration of AI tools in mental health care, the importance of safety guidelines, and the potential for AI to complement professional support.