Hi friend,
We had interesting funding rounds from Salma, nyra, and BrainCheck this week. Earnings announcements from Lifestance, Acadia and BetterHelp also gave us insights on how the mental health market is trending in 2026. And finally, there was plenty of thought-provoking new research to keep the nerds amongst us interested.
So let’s get into it.
What happened in mental health last week?
Business News:
⏺️ Salma Health came out of stealth with an $80M Series A. Salma are building a vertically integrated model that puts advanced diagnostics, rapid-acting interventions (think TMS, SAINT, Spravato) and care coordination under one roof. This is a strong signal for the growth of interventional psychiatry and to be honest, they seem like a pretty interesting business that I’m excited to learn more about. (link)
⏺️ BrainCheck raised $13M Series A for cognitive care. (link)
⏺️ nyra health raised €20M Series A for their digital neurotherapy solutions. The Austrian-based startup provides an app-based product focused primarily on neurological conditions. (link)
⏺️ LifeStance reached profitability for the first time. 2025 revenue grew by 14% to $1.42B driven by clinician network growth (+9%) and higher productivity. Net income was $9.7M reversing a $57.4M loss in 2024. They are guiding 15% revenue growth for 2026. (link)
⏺️ BetterHelp’s decline continues. BetterHelp revenue declined by 9% in 2025 as the company continues its transition from cash-pay to insurance-based reimbursement. Teladoc projects that BetterHelp revenue will decline further in 2026 before recovering as insurance volumes scale. This week’s Hemingway Report will cover all the insights from the 2025 earnings so keep an eye out for it on Thursday. (link)
⏺️ Acadia Healthcare posted mixed Q4 results. 2025 revenue grew 5% to $3.31B. But they had a MASSIVE impairment charge of over $1B which drove a headline annual loss of $1.1B. They will slow the pace of new facility openings and focus on operational improvements in the near term future. (link)
⏺️ UCLA Health received a $100M gift for mental health. Stewart and Lynda Resnick gifted $100M to UCLA Health to advance the organisations mental health work. (link)
⏺️ Kaiser therapists authorised a one-day strike. 2,400 mental health professionals at Kaiser Permanente voted 92% in favour of a one-day strike - the core dispute is Kaiser's replacement of therapist-led triage with AI and telephone operators, which clinicians say is leading to missed urgent cases. This follows a $28.3M federal parity settlement and fines in Washington and Georgia for mental health access violations. (link)
⏺️ Oregon's AI companion guardrail bill cleared the Senate. Senate Bill 1546 would require AI companions to identify themselves as software, implement evidence-based protocols for detecting suicidal ideation, and direct at-risk users to the 988 hotline - with additional restrictions for interactions with minors. State bills continue to mount in this area. (link)
⏺️ Wonder Sciences launched WonderMate, an “AI therapist” that supports patients between sessions. Welcome to the fray WonderMate…. (link)
⏺️ Amae Health opened a new Brooklyn clinic. The SMI-focused provider are one of my companies to watch in 2026 so I’m excited to see their expansion. (link)
⏺️ EvolvedMD published a white paper on the impact of its collaborative care model. Analysis of 1.3 million patients over two years found EvolvedMD's in-person Collaborative Care Model delivered 15% lower risk-adjusted total cost of care and 15% fewer psychiatric professional visits compared to traditional behavioural health models. (link)
⏺️ SEAMEO launched a $6.5M Southeast Asia health funding call. Awards of up to $6.5M are available for research projects on mental health in the area. The deadline is 24 March 2026. (link)
Research News:
⏺️ Critical safety gaps found in AI therapy. Researchers built a simulation where one AI played therapist and another played patient and then engaged in therapy sessions over the course of a simulated week. A dynamic psychological model tracked things like hopelessness and self-esteem of the patient and adjusted the AI patients responses accordingly. The study had two interesting findings. First, AIs could get pulled into patients' metaphors and essentially co-ruminate with them - sometimes even ending in simulated suicide. Second, giving ChatGPT a detailed therapist prompt actually made it less safe, seemingly suppressing its normal safety reflexes. The researchers concluded that standard AI safety testing is wholly inadequate for mental health applications. (link)
⏺️ Nature Medicine stress-tested ChatGPT Health on mental health crises. Researchers found that ChatGPT Health failed to reliably recognise mental health crises or connect users with crisis resources. Crisis alerts appeared more reliably for lower-risk scenarios than for high-risk scenarios - which, of course, is the opposite of what we would want. (link)
⏺️ NLP identified more than 2.5x more ED suicide presentations than standard coding. Australian researchers built a machine learning tool that reads emergency department triage notes and automatically identifies when a patient presented due to suicide or self-harm. They compared it against the existing standard method - manual disease classification coding - and found that the traditional coding approach had been capturing fewer than 40% of relevant presentations. Over seven years of data, standard coding identified around 10,000 cases while the new AI tool found more than 27,000, with strong accuracy. The conclusion is that this health system may have been dramatically underestimating the scale of suicide and self-harm presentations, which could have obvious knock-on effects to policy and resource allocation. (link)
⏺️ Beckley Psytech released Phase 2 results for intranasal 5-MeO-DMT. The trial showed rapid and sustained depression improvements across 12 treatment-resistant patients over 12 weeks. (link)
⏺️ Smartphone app-switching patterns linked to depressive symptoms. Researchers found that how users move across apps within a session correlates with depressive symptoms and attentional control. Users with higher depressive symptoms transitioned to a wider variety of apps. (link)
⏺️ ML model developed for risk stratification of inpatient mental health costs. Researchers tracked 3,400 Chinese psychiatric patients for three years after their first admission to see if they could predict total hospitalisation costs from information available on day one. Using machine learning, they could explain about a third of cost variation from baseline data alone. That’s not a lot but still helpful for early risk screening and planning . When they added information about what actually happened over those three years (whether patients had frequent short stays, long continuous admissions, etc.), predictive accuracy doubled. (link)
That’s all for this week. If you’d like to discuss this news with like-minded peers, consider becoming a Hemingway Pro Member. You will join a community of over 360 mental health leaders and get access to exclusive content. You can learn more here.
Keep fighting the good fight!
Steve Duke
Founder of Hemingway
