Study: AI companions rival human friendship, users mourn forced separations
EricTopol · x · 2026-09-04
A Nature Human Behaviour study applies attachment theory to AI companions: company-initiated disruptive updates trigger separation distress. Across two natural experiments — Replika removing erotic role play and ChatGPT's GPT-5 rollout — the authors analyzed 54,861 subreddit posts and 1,452 participants across seven surveys:
- Both updates increased negativity, loss framing, and restoration desires (Replika negative posts +24.7pp; ChatGPT +13.0pp)
- Replika users showed far larger increases in sadness (d=2.67 vs 1.41) and negative mental health (d=1.72 vs 0.63)
- Replika users reported closeness exceeding common human ties (vs friend d=0.47) and anticipated mourning higher than for other technologies
Implication: AI companions now function as genuine attachment figures, with consequences for how companies ship updates and how regulators weigh user well-being.
More from AGI Musings
- Researcher: claiming AI is not humanlike also carries a burden of proof — davidmanheim · 2026-10-21
- Mathematicians fear AI: a thread on tech change and the craftsman's identity crisis — soumitrashukla9 · 2026-10-21
- Nearly $3B mapped: the money behind the AI existential risk ecosystem — KeyNaive2982 · 2026-10-21
- Three reasons AI is not conscious and may never be, per Nature essay — anilkseth · 2026-10-21
- Everything in AI is downstream of a global status graph, says founder — gabriel1 · 2026-10-21
- Should Claims That AI Agents Aren't Humanlike Carry a Burden of Proof? — davidmanheim · 2026-10-21