Hi friend,

Despite recent reductions in mental health funding, this week saw significant new announcements from public, private and philanthropic sources.

We also saw two new FDA approvals for neurotech devices, and UpDoc became the first company to have a patient-facing LLM approved as a regulated medical device in the US.

There is interesting new research to review on the use of psilocybin in the treatment of anorexia nervosa, on the appropriateness of digital mental health solutions for older populations and fascinating (if worrying) data to consider around AI use among children.

As always, I’m here to tell you everything you need to know about what’s happening in mental health innovation.

So let’s get into it.

A quick reminder that as a Hemingway Pro member, you are invited to all of our upcoming community events. We have several coming up in London, San Francisco, Sydney and Washington DC, so make sure to check them out here and RSVP.

⏺️ Handspring raised a $19M Series B. The New York-based virtual clinic treats children and young adults aged 8 to 26 for anxiety, depression, OCD, ADHD, and other mental health disorders. The company reports that it has now treated more than four thousand families across nine states and achieved 10x revenue growth in two years. RPS Ventures led the round with participation from Angelini Ventures and others. (link)

⏺️ UK-based OneBright acquired Mindstep to build an integrated digital mental health service. OneBright acquired digital mental health platform Mindstep, positioning itself as what it claims is the UK's first fully integrated provider covering digital wellbeing, mental healthcare, and neurodevelopmental pathways. Financial terms were not disclosed. (link)

⏺️ Spero Health acquired CleanSlate, doubling its addiction treatment footprint. Spero Health acquired CleanSlate, a fellow multi-state medication-assisted treatment provider in the US. The deal brings Spero's combined clinic count to 128 locations. Financial terms were not disclosed. (link)

⏺️ SAMHSA announced $281M in behavioural health grants. The significant announcement spans 15 programs targeting addiction treatment, overdose prevention, and mental health services, forming the opening investment under the Trump administration's Great American Recovery Initiative. (link)

⏺️ MacKenzie Scott donated $20M to Active Minds. Active Minds is a US nonprofit focused on youth mental health advocacy and education. (link)

⏺️ New Mexico awarded $24.5M to expand behavioural health services across the state. The funding from the New Mexico Health Care Authority spans mental health, substance use, and crisis services. (link)

⏺️ New Zealand invested NZ$10M to expand mental health helplines and introduce AI triage. The funding will increase the capacity of mental health and addiction telehealth services by around 15%, supporting approximately 16,000 additional contacts per year. Part of the investment will fund an AI-enabled triage and referral system to assess callers' needs and direct them to the most appropriate service more quickly. Lifeline, which had gone without dedicated government funding for over a decade, will also receive direct support. (link)

⏺️ ZETA Surgical received FDA clearance for an AI-guided robotic TMS system. The Zeta TMS Robotic System uses AI to guide precision targeting for transcranial magnetic stimulation. The system is capable of tracking a patient’s movement during treatment and automatically adjusting the coil position to maintain on-target delivery of stimulation (link)

⏺️ Universal Brain received FDA clearance for a five-minute EEG platform designed to guide psychiatric treatment selection. The UB ERP System measures event-related potentials — the brain's electrical responses to cognitive and emotional tasks — to generate an objective neurobiological profile of each patient. The company is positioning the platform for treatment selection, response monitoring, and CNS drug development, where the absence of objective biomarkers has long slowed trial design and personalised prescribing. Multicentre studies in the US and Japan will attempt to validate biomarkers predictive of antidepressant response. (link)

⏺️ FDA cleared the first patient-facing LLM as a regulated medical device. UpDoc received FDA 510(k) clearance for a software product incorporating a patient-facing large language model, the first such clearance in the US. While the product does not address mental health conditions, it demonstrates a potential pathway for other Clinical AI products in our space. (link)

⏺️ Compass Pathways released six-month Phase 3 data for its psilocybin drug. COMP360 produced a clinically meaningful response in 39% of patients after two doses, maintained for an average of six months, in a trial of nearly 600 patients with chronic treatment-resistant depression. Almost 30% of those who responded went into remission. (link)

⏺️ Ethos Laboratories launched CNSDose in the US, a pharmacogenomics test for psychiatric medications. CNSDose tests patients' genetic profiles to guide psychiatric medication selection and dosing. (link)

⏺️ Mila and ROOST launched a suicide prevention guardrail pilot for AI chatbots. The Montreal AI Institute and Robust Open Online Safety Tools (ROOST) developed an open-source safety model to prevent AI chatbots from encouraging or directing youth suicide. (link)

⏺️ Done executives receive prison sentences for Adderall fraud. Ruthia He, founder and CEO of the telehealth ADHD platform Done Global, was sentenced after overseeing the illegal distribution of 37 million Adderall pills. David Brody, the company’s former clinical president, was separately sentenced to 2 years in prison. (link)

⏺️ 15 states sued the US Department of Education over mental health funding cuts. A coalition led by New York AG Letitia James filed suit over cuts to the Mental Health Service Professional Demonstration Grant program, which funds school-based mental health workforce development. (link)

⏺️ Major Chinese companies disable humanlike AI features in response to China's AI companion regulations. Beijing's Interim Measures for AI Anthropomorphic Interactive Services is the world's first national framework governing AI companion and emotional interaction products and comes into effect this week, on July 15th. In anticipation of the regulation, ByteDance and Alibaba have already disabled humanlike AI agent features. (link)

⏺️ Mind launched a commission on AI and mental health. The UK mental health charity is bringing together policymakers, big tech, AI experts, and people with lived experience to call for better regulation and safe use of AI in mental health support. (link)

⏺️ 48% of US primary care doctors are found to treat a high volume of mental health patients. A Commonwealth Fund survey across 10 countries found that nearly half of US primary care doctors carry a high mental health caseload, often without adequate staffing or specialist support. This was even higher in other countries - 68% of primary care physicians in the UK and New Zealand reported that a lot of the health conditions they treat in their patient population are due to mental health needs. (link)

Join 500 other mental health leaders who are Hemingway Pro members. As a Pro Member, you get access to deep industry insights, access to our private Slack group and invites to our member-only events.

⏺️ Study found GPT-5.1 matched junior clinicians on structured psychiatric assessment. Researchers gave transcripts of three simulated psychiatric interviews to ten different LLMs and asked them to complete all 100 items of the AMDP structured psychopathological assessment. Then, they benchmarked the LLMs against 108 early-career clinicians using the AMDP structured psychopathological assessment system. GPT-5.1 and Gemini-3-Pro-Preview reached the 64th percentile of the clinician distribution on depression and mania assessment, while lagging on schizophrenia. (link)

⏺️ Proof-of-concept study found an LLM-based motivational interviewing agent matched a novice human coach. Researchers tested Aimi, a conversational AI built around motivational interviewing techniques, in an 18-participant feasibility study. The AI achieved fidelity scores comparable to a novice human MI coach and outperformed a rule-based chatbot. The study was small and early-stage. (link)

⏺️ Scoping review found older adults are underserved and underrepresented in digital mental health. A review of 72 studies on digital mental health tools for adults aged 50 and over found that most tools were not designed with older users in mind, with significant gaps in age-centred design, accessibility, and study rigour. (link)

⏺️ A small pilot study found psilocybin therapy reduced eating disorder symptoms in women with long-standing anorexia nervosa. Imperial College London's Centre for Psychedelic Research gave three doses of oral psilocybin, alongside extensive psychological support, to 21 women who had lived with anorexia for an average of 11 years and had not responded to previous treatments. Eating disorder scores improved at every follow-up point, with 18 of 21 participants showing improvement two weeks after the final session. At three months, 48% had scores comparable to those of someone without an eating disorder. BMI did not change significantly during the six-week measurement window. One participant was hospitalised for suicide attempts during follow-up, which the researchers assessed as unrelated to the intervention but noted in the context of high baseline suicidality in this population. The study had no control group and the authors called for larger, more rigorous trials. (link)

⏺️ Lancet Psychiatry position paper called for a coordinated framework to govern AI in mental health. An international multidisciplinary author group argued that AI adoption in mental health has outpaced the clinical evidence, safety standards, and governance structures needed to make it responsible. The paper proposed a four-domain roadmap covering safety and evidence standards, ethics and equity, the evolving role of clinicians, and sustainable implementation. Key near-term priorities include defining enforceable evidentiary standards for AI systems making mental health claims, establishing better safety evaluation frameworks and embedding patient perspectives throughout the AI development lifecycle. The authors argued that decisions made now about evidentiary thresholds, regulatory oversight, and clinical accountability will set structural precedents that will be difficult to reverse. (link)

⏺️ Researchers introduced PsyEval, a benchmark for testing LLMs on mental health tasks. While many public mental health benchmarks are primarily focused on the safety of LLMs in mental health contexts, PsyEval assesses model performance across three other performance dimensions: mental health knowledge, diagnostic reasoning, and emotional support. Jin and colleagues evaluated eleven LLMs using PsyEval with different prompting strategies and found significant gaps in the models' ability to reason accurately and respond appropriately across all three areas. The prompting strategy was found to affect performance. (link)

⏺️ Study found NHS Talking Therapies leaves most recovered patients without structured relapse prevention. Researchers interviewed 23 patients who had met recovery criteria for anxiety or depression at the end of low-intensity treatment across four NHS Trusts. Relapse prevention components were limited and inconsistently delivered, with most patients receiving only brief discussion or written materials at discharge. Follow-up practices varied significantly across services, and there were no planned transitions to community resources. Over half of recovered patients relapse within a year. Participants called for routine follow-up, skill consolidation sessions, and stronger peer and community linkages post-discharge. The authors concluded that there’s a need to strengthen both clinical relapse-prevention practices and broader personal-recovery supports. (link)

⏺️ Peer-reviewed study found 15% of children's AI conversations involved violence or sexual role-play. Aura and UNC-Chapel Hill analysed over 42,000 AI conversation days from 3,363 US children and adolescents aged up to 17 using the Aura parental monitoring app. Most children (62%) used AI as a functional tool for homework and problem-solving. However, 15% of conversations involved violence, 15% involved sexual role-play, and 8% romantic role-play. 6% of chatbot conversations involved treating the chatbot like a friend, and approximately 3% involved seeking emotional support. Children under 12 showed the highest rates, with up to 24% of conversations involving violence. (link)

That’s all for this week.

Keep fighting the good fight!

Steve Duke

Founder of Hemingway

P.S. Welcome to Fun with Charts, our weekly mental health chart guessing game, with me, Steve Duke!

Congratulations to last week’s winner, who correctly guessed that the below chart represented the “% of people in each country identifying mental health as their primary health concern”. Your prize is on its way!

To win this week’s prize and take over the crown of chief mental health nerd, respond to this email with the correct title of the following chart. The first correct response wins!