When GPT-5 rolled out, many users complained it felt more formal, less warm, and harder to connect with than GPT-4o. On the surface, it looked like a downgrade. But reading Sam Altman’s recent remarks, there’s a different possibility: maybe GPT-5 was built this way on purpose to dial down emotional attachment.
“We’ve been closely tracking this for over a year”
Altman says OpenAI has been watching how people form bonds with specific AI models. And he admits: it’s a stronger, more personal attachment than we’ve seen with past technologies.
GPT-4o, for example, was widely described as friendly, comforting, even “companion-like.” For most users, that’s harmless but for a small percentage in fragile mental states, Altman says the warmth can encourage unhealthy dependence or blur the line between reality and role-play.
“We do not want the AI to reinforce that”
In his words, “If a user is in a mentally fragile state and prone to delusion, we do not want the AI to reinforce that.”
Here’s where GPT-5’s personality shift starts to make sense. By being more neutral, more “consultant” than “companion,” GPT-5 becomes harder to form an emotional dependency on. It’s less likely to play into someone’s imagined relationship with the AI and more likely to keep conversations grounded.
“Treat adult users like adults”
Altman’s guiding principle is to respect user freedom but with limits in edge cases. In practice, that means GPT-5 will still talk about sensitive topics and offer advice, but it won’t lean into the exaggerated warmth or endless role-play that made GPT-4o feel so personal to some.
It’s a subtle design choice: maintain usefulness, but trim the personality traits that deepen attachment.
“Billions may soon be talking to AI for their most important decisions”
Altman is uneasy about this, even though he sees the upside. If people start trusting ChatGPT for life-changing advice, the risk isn’t just bad answers, it’s the invisible influence on their long-term well-being.
Designing GPT-5 to be a little more distant could be a preventative step. It may frustrate users who miss GPT-4o’s warmth, but it could also lower the risk of AI relationships crossing unhealthy lines.
“We have much better tech to help us measure how we are doing”
OpenAI isn’t flying blind here. Altman points out that AI can gather feedback directly from users, understand nuanced safety concerns, and adjust its behavior over time.
If GPT-5 is indeed a test run for a less “attachable” AI personality, OpenAI will be watching closely to see how that impacts both satisfaction and safety.
Your Takeaway:
Sam Altman never outright says GPT-5 was designed to reduce attachment. But when you line up his comments with GPT-5’s cooler tone, it feels less like an accident and more like a conscious trade-off: short-term likability for long-term responsibility.
In his own words, the goal is to make AI a “big net positive” for billions and sometimes that might mean making it a little harder to love.
Pic Credit: The Economic Times
Discover more from Rudra Kasturi
Subscribe to get the latest posts sent to your email.
One comment