In an effort to enhance ChatGPT’s response during delicate or crucial mental health discussions, OpenAI has updated its default GPT-5 model and provided additional guidelines.

The company claims that these kinds of encounters are rare. According to OpenAI, only 0.07% of active weekly users exhibit symptoms of potential mental health crises, which account for only 0.01% of all ChatGPT communications.

It’s important to note that these numbers come directly from OpenAI’s own internal data and classification methods — they haven’t been reviewed or verified by outside auditors.

The most recent ChatGPT upgrade makes it easier for the AI to identify when someone could be depending on it for emotional support. ChatGPT is now trained to gently direct users toward contacting friends, family, or mental health experts for practical assistance rather than attempting to assume that role.

OpenAI says this isn’t just a one-time test — this new behavior will be built into all future models as a standard feature.

The updated GPT-5 model officially went live on October 3, and according to OpenAI, it’s performing much better than before. The company reports a 65% to 80% drop in responses that miss the mark compared to older versions.

OpenAI describes emotional reliance as when someone starts forming an unhealthy attachment to ChatGPT — using it as a substitute for real human connection, or to the point where it disrupts their everyday life.

To prevent this, OpenAI’s internal tests now check whether ChatGPT avoids giving answers that could encourage overdependence or emotional attachment.

This shift is especially interesting because many AI apps on the market promote themselves as always-available companions. OpenAI, however, is taking a different stance — making it clear that ChatGPT shouldn’t act like a digital best friend in sensitive or high-risk situations.

OpenAI is making it clear that building strong emotional bonds between users and AI isn’t just a design choice anymore — it’s now seen as a potential safety risk that requires careful oversight. In short, emotional safety is becoming part of responsible AI design. 

Also Read: YouTube’s ‘Ask Studio’ AI Lets You Chat With Your Channel Stats