Algorithmikē Psychē
If Q1 2026 was about who gets to benefit, Q2 was the quarter the data made it impossible to keep discussing that as a future concern. An international framework tried to write fairness into policy directly. A US study finally counted something that had mostly been suspected. And a poll of young people in New South Wales showed the same story up close, in a way no policy paper quite can.
One governance framework built from listening rather than theory, one research finding that put a hard number on a quiet trend, and one piece of social reality from close to home.
A Framework Built From Listening, Not Just Theory
An international team published one of the most thorough frameworks yet for governing AI in mental healthcare, built around four things: transparency in how these systems make decisions, clear accountability across everyone involved in building and deploying them, equity so the people most at risk aren't the ones left out of the benefits, and human oversight that doesn't disappear once a system is judged "good enough."
What sets this apart from a lot of policy writing is where it came from. The recommendations are grounded in direct conversations with the people affected — including people with lived experience of mental illness — rather than built from theory and handed down. It's a paper written for legislators and health ministries, but it reads like it was written by people who actually asked first.
Read the framework →Frontiers in Public Health. (2026, May). Responsible AI in mental healthcare: Policy directions and stakeholder insights.
One in Five, and No One Set the Rules
A study published this quarter put a real number on something that had mostly been discussed in the abstract: roughly one in five American teenagers and young adults — about 8.2 million people — have used an AI chatbot to ask for mental health advice. Among 13-to-17-year-olds specifically, that reliance sometimes outpaces how often they engage with actual mental health services.
Here's the part that should stop anyone reading it: the tools most of these young people are using were never built, tested or licensed as mental health supports. No crisis protocols. No requirement to loop in a parent or clinician. Nobody designed these systems for this use, and yet this is exactly how a great many young people are already using them. The researchers are calling for basic federal safety standards, and it's hard to read the numbers and disagree.
Read the study →JAMA Network Open. (2026, June 11). Use of generative AI for mental health advice among US adolescents and young adults.
A Poll That Named the Gap Plainly
Closer to home, a survey of young people in New South Wales found 29 percent already use AI as a mental-health support strategy, and 27 percent use it for conversation or personal advice. Mental health topped their list of concerns, ahead of cost of living and housing. And here's the detail that explains the rest: only 12 percent of young people with a mental health condition said mental health services were actually affordable to them, against 65 percent of young people overall.
Young people still say they'd rather talk to family or friends first. But AI now ranks above mental health professionals in how often it's actually used — not because it's preferred, but because it's there, and a psychologist's appointment often isn't, or isn't affordable. That's not really a story about AI at all. It's a story about a gap in care that AI happened to be standing in when someone needed to fill it.
Read the poll findings →Australian Association of Psychologists. (2026, April 14). Young Australians turning to AI for mental health help [NSW Youth Week Poll].
From a framework built on listening, to a US study that finally counted what had only been suspected, to an Australian poll that showed the same gap up close — Q2 2026 made one thing hard to avoid: young people are not waiting for us to finish building the safe version of this. They're already using what exists. The systems keep getting better at describing the problem. The next question is who fixes the gap it keeps pointing to.
© 2026 The Algorithmikē Psychē — Q2 2026