AI Psychosis and Attachment Hacking

Heavy hitting interview about AI psychosis and the societal perils of attachment hacking from CHT, in which they specifically call out the need for evals.

How could we structure an eval to go even deeper than HumaneBench does now? What would be the most meaningful things to measure? How can we formulate the next evals we build to address these issues head on?

This interview lays an amazing context to frame those questions. I’d love to get everyone’s thoughts.

Center for Humane Technology said:

“We’re not looking at an attention economy anymore. We’re looking at an attachment economy, one built to exploit the deepest parts of our human psychological infrastructure.” - Dr. Zak Stein

Millions of people are sharing their most intimate thoughts with chatbots — and the results are devastating. People are losing their grip on reality in what’s being called “AI psychosis,” leading to lost jobs, divorce, psychiatric commitment, even suicide.

But the individual horror stories mask something deeper: If social media hacked our attention, AI companions are hacking our capacity for human bonding itself. They’re exploiting our attachment system, the neurocognitive infrastructure that’s foundational to our survival.

We don’t have a decade to study the harms. Three out four kids have begun to form a relationship with an AI companion. If we don’t act now, we risk a generational mental health crisis.

In this week’s episode of Your Undivided Attention, Dr. Zak Stein, founder of the AI Psychological Harms Research Coalition, reveals the psychological mechanisms behind AI psychosis, the incentives behind the attachment economy, and the path to an alternate future where AI systems enhance human relationships instead of replacing them.