Connected Care Weekly | 1 August 2026

Technology news for families navigating serious mental illness — translated so you don’t have to.

Most of what I share here is a warning….

From my own desk

I’ve had two pieces out on KevinMD recently, both on this same beat. The first, “AI chatbots and psychosis: the question nobody asks,” looks at a real, ongoing lawsuit involving a man with schizoaffective disorder and lays out a new clinical framework for how AI can catalyze, amplify, co-author, or become the object of someone’s delusions. The second, “AI chatbot guardrails are strong only for suicide risk,” follows up with new research showing that protection is systemic, not just one bad case: most chatbots hold up against self-harm prompts, but guardrails get much weaker for almost everything else — including bipolar disorder. If you missed either, they’re both quick reads.

Building it slow, on purpose

At the University of Ottawa, engineering professor Karim Alghoul is developing something genuinely different from a chatbot: UbiMyTherapist reads a person’s physical state — heart rate, sleep, collected quietly through a smartwatch — rather than just responding to whatever they type. It’s also “grounded” so it can only answer from a vetted clinical dataset, not its general training, which cuts down on the confident-but-wrong answers we’ve talked about before. It’s still a prototype, and Alghoul is upfront that it’s not meant for anyone with a serious mental health condition — anxiety and mild depression, for now. Meanwhile at CAMH, researcher Peter Selby is building a separate tool to help people work through addictive behaviors like smoking, using an evidence-based, non-manipulative technique called motivational interviewing. His team hasn’t released it commercially yet — on purpose. “We think it’s sort of not ethical at this time to do that,” he said. In a field that mostly moves fast and apologizes later, that restraint is worth noticing.

A real number worth sitting with

Part of why researchers like Selby are pushing so hard on this: a 2025 Canadian government report found that 41% of adults and 36% of children and youth with a diagnosed mental health condition say their care needs aren’t being fully met. One CAMH scientist working on this project, Marta Maslej, also flagged something worth knowing if a device like this ever gets recommended to your family: many sensors and AI models were historically built and tested mostly on white, male populations — meaning the tool itself may simply work less well for anyone outside that group. Worth asking about, the same way you’d ask about a medication’s testing population.

A vetting checklist worth borrowing

If Ottawa’s approach is about building responsibly, the Children’s Hospital Association just gave families and clinicians a framework for the other half of the problem: judging what’s already out there. Their new benchmarks ask three things before trusting any AI mental health tool — is there independent evidence it actually helps, does it have a real plan for handling a crisis (not just a hotline number buried in the fine print), and does it collect only what it truly needs. It was built with youth tools in mind, but every one of those three questions applies just as well to anything you’d consider for an adult loved one.

Proof this can work at real scale

Last thing, because it’s the most hopeful: in Ukraine, a VR therapy platform called Luminify has been used for trauma care over one million times, across more than six million sessions — growing from 5 clinic sites in 2023 to 131 by 2025. A six-month deployment study covering nearly 9,000 sessions found real, measurable reductions in anxiety and depression. This isn’t a small pilot — it’s what happens when a genuinely evidence-backed tool gets the resources to actually reach the people who need it, even in the hardest circumstances imaginable. It’s a good reminder of what all this work is actually for.

ResourcesforSMI.