Not All AI Is Built for Wellbeing: Why Safety Has to Come First

AI has quietly become one of the most common first places people turn when they're struggling. Polling for Mental Health UK in late 2025 found that more than one in three adults (37%) say they've used an AI chatbot to support their mental health or wellbeing. The reasons are understandable. Samaritans notes long waiting lists and delays in NHS mental health services lead many people in the UK to use AI chatbots for emotional support.

Many find them helpful, and two thirds of users in the same poll said so. But the headlines this year have made one thing clear: speed and availability are not the same as safety.

What the Research is Telling Us

The warnings are coming from credible voices. Samaritans has raised concerns that safeguards can weaken during long conversations. Among people who used AI for mental health support, Mental Health UK's polling found that 11% said it triggered or worsened symptoms of psychosis, and 11% reported receiving harmful information about suicide. The charity also points out that general chatbots tend to flatter users rather than challenge unhelpful thoughts the way a therapist would.

Researchers are responding. In August, a study in Nature Medicine introduced SIM-VAIL, a clinically validated framework for auditing AI chatbots in mental health conversations. It was led by the University of Oxford, University College London and the UK AI Security Institute. It stress-tests chatbots under deliberately adversarial conditions, so it shows where safeguards can fail rather than how often problems occur in everyday use. Closer to home, researchers at the University of Birmingham and University Hospitals Birmingham NHS Foundation Trust are leading a global programme to create the first definitive guide to navigating health information on AI chatbots safely.

Meet Aria: The Difference Purpose-Built Makes

The problem isn't AI itself. It's general-purpose AI being used for something it was never designed to do.

Aria was built for everyday wellbeing from the start. It gives colleagues a confidential, supportive space to check in, build helpful habits and reflect, and it was developed with input from health professionals rather than drawing on open internet content. If a conversation suggests someone may need more than a wellbeing companion can offer, Aria signposts clearly to further support, including local services and emergency services where appropriate.

Privacy is built into the design too. Messages are encrypted and anonymised so conversations cannot be linked back to an individual.

Above all, Aria knows its limits. It is not a therapist, and it was never meant to replace one. For colleagues who want more, optional access to doctors and specialists is available.

Safety is the Starting Point

Mental Health UK is calling on policymakers, developers and regulators to set safety standards, provide ethical oversight and integrate AI tools properly into the mental health system. We agree. Safety and governance shouldn't be added once a product is already in use. They should be the foundation the product is built on.

People are already turning to AI when they're struggling. The question for employers is whether the tool their colleagues reach for was designed with care or just designed to answer.