HAI policy workshop held in June 2026 has surfaced something many of us feel intuitively: the mental health AI tools we increasingly turn to when therapy feels out of reach have outpaced the guardrails designed to keep us safe. Bringing together researchers, clinicians, policymakers, and patient advocates, the convening identified critical gaps in how we regulate the chatbots, CBT apps, and companion bots like Character.ai and Replika that now fill a void traditional care cannot. For anyone who has ever typed a worry into a screen at 2 a.m., this conversation is not abstract.
The Paradox We Are All Living Inside
We exist in a quiet tension right now. The same tools that promise greater access and lower cost can also deliver substandard care when no clear standards exist. Stanford HAI's workshop flagged how headlines have surfaced again and again: minors forming unhealthy emotional attachments to chatbots, users in crisis receiving harmful or inadequate responses, and research showing that general-purpose AI commonly misses warning signs of distress. The promise is real, and so is the danger.
Why the Response Cannot Keep Pace
What the workshop underscored is just how fragmented the policy landscape has become. More than 140 state bills related to AI in mental health have been introduced, yet federal legislation remains pending and narrowly focused on protecting minors. Meanwhile, the evidence base lags behind: many purpose-built mental health tools lack validated outcomes and representative samples, and the models powering general-purpose chatbots update so rapidly that yesterday's safety research quickly becomes outdated. We are, in essence, trying to anchor a moving ship.
How We Move Forward With This in Mind
If you have been using—or considering—an AI tool for emotional support, this is not a moment to feel shame about reaching for help. It is a moment to anchor your choices in a few grounded questions. Ask whether the tool has been evaluated with rigorous study designs, whether the company behind it has published safety data, and whether a licensed clinician is involved in your care. We can still benefit from what AI offers—accessibility, immediate support, a bridge during long waits—while recognizing that our wellbeing deserves more than a product with no validated outcomes.
One small step to take this week: before opening a chatbot tonight, pause and write down what you are actually looking for. A listening ear? CBT exercises? Crisis support? Naming the need is not a trivial thing. It is the first act of advocating for the kind of care that genuinely serves you.