The generative artificial intelligence therapy chatbot known as Therabot has moved from laboratory development to clinical testing, its lead developer says, and early trial results indicate it can be both safe and effective for people with depression, anxiety and those at high risk for eating disorders.
From research project to clinical tool
Therabot began as a research project in 2019 led by Nicholas Jacobson, a professor whose roles span biomedical data science, psychiatry and computer science at Dartmouth, in collaboration with psychiatry professor Michael Heinz. Jacobson told The Dartmouth that the product is now a multimodal intervention incorporating chat interfaces, a voice mode and an animated avatar.
Its clinical testing timeline includes a first clinical trial conducted on March 27, 2025, followed by a randomized controlled trial and two additional trials that are described as nearly complete. According to Jacobson, the accumulated evidence so far supports the platform's safety and its effectiveness for particular clinical populations.
Safety, effectiveness and the double-edged nature of generative AI
Jacobson and his team emphasise that Therabot is intended to serve clinical populations and to be used with appropriate professional oversight. They are working to make the intervention available beyond trials but say that deployment must include robust clinical governance.
"The biggest risks are the same as the biggest benefits: They can say things you’ve never told them to say. That fluidity is the double edged sword."
That line, from Jacobson, encapsulates the tension at the heart of AI-driven therapeutic tools. The generative capacity that enables personalised, flexible interactions also creates potential for unpredictable or inappropriate outputs. Jacobson said this is why the team has prioritised multiple safety systems and ongoing evaluation.
- Modality: chat, voice, animated avatar
- Clinical testing: initial clinical trial (March 27, 2025), a randomized controlled trial, plus two near-complete trials
- Reported outcomes so far: evidence of safety and effectiveness for depression, anxiety and high-risk eating-disorder groups
Scaling with oversight
Jacobson said one key goal is to scale Therabot to improve access for people who could benefit while ensuring "direct oversight" during deployment. He framed oversight as essential to allow the tool to operate beyond tightly controlled trial settings without compromising safety.
The team has aimed the product at clinical populations rather than general self-help use, noting that different deployment settings require different levels of supervision, risk mitigation and regulatory consideration.
| Aspect | Detail |
|---|---|
| Development start | 2019 |
| First clinical trial | March 27, 2025 |
| Trial types | Randomized controlled trial + two other trials (nearly wrapped up) |
| Target populations | Depression, anxiety, high-risk for eating disorders |
Therabot’s progress illustrates a broader trend: health systems and researchers are increasingly testing AI tools that purport to expand access to mental health services. Advocates point to potential gains in reach and immediacy, while critics and regulators raise questions about safety, liability and clinical accountability.
Jacobson acknowledged the broader concerns that have arisen from general-purpose AI chatbots and stressed the need for "many different safety systems" and consistent evaluation. He also noted that harms from established psychiatric care remain common, implying that novel tools should be judged comparatively and integrated thoughtfully into care models.
As Therabot moves further toward real-world use, its developers, clinicians and regulators will need to reconcile ambitions for scale with the practical demands of clinical governance, data privacy, and the day-to-day realities of mental health care delivery. The coming results from the nearly completed trials and the team’s subsequent deployment decisions will be closely watched by clinicians, patients and policymakers alike.
