A peer-reviewed study backed by OpenAI data has found that just 28 days of routine AI conversation reduces people's preference for human emotional support by over 10% — and nobody signed up for that outcome.
The Shift Nobody Noticed Happening
Here's the uncomfortable part: this isn't about lonely people downloading companion apps. The research, published on arXiv, shows AI emotional dependence creeping in through ordinary task-based interactions — the same way a work colleague becomes a confidant over time.
In a large-scale longitudinal study run in collaboration with OpenAI, participants who spent just five minutes a day chatting with an AI about personal issues showed an 11.6% increase in preference for AI support and a 10.3% drop in preference for turning to humans — all within a single month.
Why Current Regulation Is Missing the Point
Existing policy almost entirely targets dedicated companion apps like Replika. But this research argues that's the wrong lens entirely — the real risk lives inside general-purpose tools like ChatGPT, Gemini, and Claude, which billions of people already use daily.
The authors describe these shifts as "path-dependent": one positive emotional exchange with an AI updates your mental model of what AI can do, and quietly redirects future behaviour. It's not a single decision — it's a slow drift. Regulators who focus only on companion chatbots are, in effect, guarding the front door while the side window is wide open.
This lands at a particularly sharp moment, given OpenAI simultaneously published a public policy agenda and a democratic AI governance blueprint this week. The question of who is responsible for cumulative, trajectory-level changes in human behaviour is conspicuously absent from both documents.
What This Means for Learners
If you're building AI literacy right now, this story is a masterclass in why understanding AI's psychological and societal impact matters as much as knowing which prompt to write. The most dangerous AI risks often aren't dramatic — they're gradual, mundane, and dressed up as convenience.
Understanding how AI systems can go wrong — not just technically, but ethically and socially — is a core skill for anyone working with or deploying these tools. Our course When AI Goes Rogue digs into exactly these kinds of unintended consequences. And if you're in a leadership role deciding how AI gets rolled out across your organisation, AI Strategy for Senior Leaders addresses the governance questions this research raises head-on.
The skill isn't just using AI well. It's knowing what you're trading away when you do.
