Hi friend,
For a change, it was a quieter week within the mental health industry. The HHS awarded $77M in new grants; the One Mind Accelerator opened applications for its 2027 cohort, and California progressed SB-903, which aims to draw clearer lines on where AI can (and mostly can’t) be used in therapy.
On the research front, however, there are many interesting papers to review.
If you think using AI might be making you dumber, new research suggests you are correct. The Mental Elf (one of my favourite publications) wrote a great summary of how the Australian teen social media ban is performing. There was an interesting paper associating better GLP-1 usage and better health outcomes among patients with severe mental illness as well as research from Denmark linking traumatic experiences and long-term cardiovascular health.
Let’s get into it!
Join our Session with Thomas Insel MD
I'm incredibly excited to have Thomas Insel MD join us for a Hemingway Session next month. Tom will be sharing the lessons he has learned from a decade in behavioural health start-ups (and several decades more working across the mental health ecosystem).
Sign up here 👉 https://luma.com/hemingway-1rv8
FYI, Sessions are a benefit of Hemingway Pro membership. To join our Sessions and access all of the other benefits of being a Hemingway Pro Member, click below.
What happened in mental health last week?
Industry News:
⏺️ In the US, HHS awarded $77M in new substance use and mental health grants. The SAMHSA funding includes $22.2M for prevention, $23.6M for treatment and recovery and $21.8M for community mental health programmes. Another $9.4M will support suicide prevention and crisis services, including $7.5M for 988 Lifeline crisis centres to follow up with people after an initial crisis contact. (link)
⏺️ The One Mind Accelerator opened applications for its 2027 cohort. Previous participants include Ampa, Amae Health, Limbic, Jimini Health, Slingshot, Flow Neuroscience, Freedom Biosciences and many more leading mental health businesses. Applications are reviewed on a rolling basis (with early submission encouraged) and close on October 19th. There is also an info session this Wednesday, which you can sign up for here. (link)
⏺️ The UBS Optimus Foundation issued a call for proposals for outcomes-based mental health projects in low- and middle-income countries. The programme will provide funding and technical support to projects using outcomes-based financing to improve mental health, with a focus on low- and middle-income countries. Priority areas include youth mental health, suicide prevention, community care, employment, education and tech-enabled care. Expressions of interest are due October 2nd. (link)
⏺️ ATAP and ERPHealth launched an outcomes-based behavioural health network for union members and their families. ATAP is a nonprofit employee assistance programme serving union workers, connecting members to mental health and substance use care. Its new High Performance Network will use ERPHealth to independently track patient engagement, clinical progress and outcomes across participating providers. The model combines measurement-based care, digital engagement, peer navigation and value-based payment, aiming to direct union members towards providers that can demonstrate better results. (link)
⏺️ Syra Health and HealthSync were selected to participate in ACCESS. The companies plan to use Syra's Syrenity behavioural health platform to support people enrolled in Medicare. (link)
⏺️ Project Healthy Minds and Harvard Business School launched a research coalition to study what works in workplace mental health. The Workforce Lab brings together founding companies including L'Oréal USA, ServiceNow, KPMG, Apollo Global Management and Universal Music Group, with Harvard Business School professor Ryan Buell as an academic collaborator. The initiative aims to build evidence on which employer mental health interventions produce meaningful outcomes. (link
⏺️ California lawmakers gave final approval to a bill restricting the use of AI chatbots as therapists. SB 903 was introduced in January and passed the Senate in May before clearing its final legislative votes last week. The bill would prohibit companies from offering or advertising therapy or psychotherapy delivered by companion chatbots, while allowing licensed clinicians to use AI for limited supplementary and administrative purposes. It would also prevent AI from independently making therapeutic decisions, detecting emotions or generating treatment plans without professional review. The legislation builds on broader chatbot safeguards California introduced last year. (link)
⏺️ A bipartisan group of US lawmakers introduced legislation to permanently extend enhanced Medicaid funding for mental health crisis services. The 988 Crisis Response Act would maintain an 85% federal matching rate for qualifying mobile crisis teams and extend it to crisis call centres and stabilisation and receiving facilities, covering the three major components of the 988 crisis-care continuum. (link)
⏺️ In England, mental health-related A&E visits among children increased 36% between 2019 and 2025. Attendances rose from 55,525 to 75,491 (+36%) compared with an 8% increase in overall A&E attendance. Around 8% of children attending with a mental health crisis, more than 6,200, waited over 12 hours. The Royal College of Paediatrics and Child Health and Mind argue that A&E is poorly suited to children in mental health crisis but inadequate community support is leaving some families with nowhere else to go. (link)
⏺️ LifeStance released real-world data showing 61% of patients with treatment-resistant depression improved meaningfully after TMS. LifeStance analysed PHQ-9 scores from 355 patients treated between January 2025 and May 2026. Among 258 patients with more severe depression at baseline, 61% achieved a clinically meaningful reduction in symptoms. The analysis had no comparison group, so it cannot show how LifeStance's outcomes compare with other treatments or TMS providers. LifeStance plans to expand TMS from 11 to 22 states by the end of 2026. (link)
⏺️ Yuna developed a structured memory architecture designed to preserve clinically relevant context across AI mental health conversations. The system separates memory into four types: episodic memory for specific experiences, pattern memory for recurring behaviours and emotional themes, semantic memory for stable personal context, and working memory for the user's current state. Memories are extracted and updated across conversations, allowing the system to distinguish enduring patterns from immediate concerns and retrieve relevant context in later interactions. (link)
Join more than 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.
Research News:
⏺️ A new paper outlined how consumer AI chatbots are becoming an unofficial entry point into healthcare. Eric Topol and colleagues considered how general-purpose AI used for health questions could interact with formal care, including how AI-generated information influences clinical decisions and how patients move between AI and human care. The authors propose evaluating these systems at the pathway level, including their impact on care completion, transparency around how patients are routed, data governance and whether patients can easily leave for another provider or platform. They highlight that deeper integration could improve access, particularly where healthcare is fragmented, while also concentrating significant power over triage and patient data in a small number of technology companies. (link)
⏺️ A new review found relying on AI can impair learning and contribute to the loss of skills that require regular practice. Writing in Trends in Cognitive Sciences, researchers reviewed evidence on “cognitive offloading” — when people hand mental tasks over to AI — and concluded that it can interfere with acquiring new knowledge and cause specific skills to deteriorate when people stop practising them. However, they found much less reason to think AI will weaken fundamental cognitive abilities such as intelligence or working memory and concluded (somewhat unsurprisingly) that the effects depend heavily on how AI is used. (link)
⏺️ Research found disruptive changes to AI companions were associated with responses resembling separation distress. Researchers analysed 54,861 Reddit posts around Replika's removal of erotic role-play and ChatGPT's rollout of GPT-5, alongside seven surveys involving 1,452 people. Negative posts increased by 24.7 percentage points after the Replika change and 13 points after the ChatGPT update, alongside more expressions of sadness, loss and desire to restore previous versions. Surveys also found Replika users could form strong attachments to their AI companions. The findings suggest product changes can have psychological consequences for highly attached users. My recent post on Tolan offers further anecdotal evidence of such dynamics. (link)
⏺️ A large Danish study found people who developed mental health problems after a traumatic event were associated with greater risk of later cardiovascular disease. Researchers compared nearly 44,000 trauma-exposed adults who later had a heart attack, stroke, heart procedure or died from cardiovascular disease with around 176,000 who did not. PTSD, depression, anxiety and substance use disorders were all linked to greater cardiovascular risk, although the strongest associations differed between men and women. The findings suggest the mental health consequences of trauma more broadly — not just PTSD — may be important when assessing someone's future cardiovascular health. Because the study was observational, it can’t show that the mental health conditions caused the cardiovascular problems. (link)
⏺️ GLP-1 drugs were associated with lower mortality and cardiovascular risk among people with serious mental illness. Researchers used electronic health records to compare people starting GLP-1 drugs with similar patients starting SGLT2 inhibitors. Over four years, among those with serious mental illness, 4.9% of GLP-1 users died over four years compared with 6.5% of SGLT2 users. They also had fewer heart attacks, strokes and other major cardiovascular events. The analysis was observational but creates further interest in GLP-1’s ability to treat mental health conditions. (link)
⏺️ Early research found Australia's under-16 social media restrictions produced little evidence of a substantial reduction in teenagers' social media use during their first three months. A blog post from the Mental Elf analysed recent research on the Australian under-16 social media ban. Researchers followed 436 young people aged 12–16 before and after the policy took effect. Although 14–15-year-olds reported some reductions in daily use and time spent on social media, more than 86% of under-16s had still accessed at least one restricted platform in the previous week. Some used other people's or fake accounts and private browsers to get around restrictions, while only 66% reported encountering an age-verification system. The findings suggest the policy had limited impact on behaviour in its first three months, although implementation was incomplete and the study relied on self-reported use. (link)
⏺️ An 11-year analysis found heavier social media use was consistently associated with depression, with the relationship becoming more apparent above roughly 2.5 hours a day. Researchers analysed 11 annual surveys from 2014–2024 involving more than 183,000 adults aged 18–70 and used machine-learning models to examine how social media use contributed to identifying people reporting depression. Social media use began contributing positively to depression classification at around 150 minutes per day. The study is notable because the association appeared repeatedly across different samples over more than a decade, despite major changes in social media during that period. However, the surveys were cross-sectional, so they cannot tell us whether heavier social media use contributes to depression, people with depression use social media more, or other factors explain the relationship. (link)
⏺️ A rare placebo-controlled study helped separate brain changes caused by early psychosis from those potentially caused by antipsychotic medication. Scientists have long known that people with psychosis often experience progressive thinning of the brain's cortical grey matter, but it has been unclear whether this is caused by the illness or antipsychotic medication. To separate the two, researchers randomly assigned 62 people experiencing their first episode of psychosis, who had never taken antipsychotics, to medication or placebo alongside intensive psychosocial treatment. Over a year, those off medication showed widespread cortical thinning compared with healthy participants, while the same widespread changes were not seen in those taking antipsychotics. This suggests the thinning is more likely related to psychosis itself than caused by medication. (link)
⏺️ Researchers identified genetic and epigenetic changes in the brain that may help explain what goes wrong in major depression. Using several molecular techniques, they analysed postmortem tissue from the hippocampus. They found that healthy adult brains continue producing new neurons, while in people with major depression this process appears to start but stall before the cells fully mature. They also found changes in how genes were regulated across pathways involved in stress, immunity, metabolism and communication between neurons. Together, the findings provide a more detailed biological picture of how the hippocampus is altered in depression and could point towards new treatment targets. (link)
⏺️ Leading psychiatrists argued that “interventional psychiatry” should become a more formally defined part of the profession as procedure-based treatments expand. The term is already widely used for treatments such as brain stimulation and ketamine, but there is no consistent definition of what an interventional psychiatrist should do or how they should be trained. Writing in Nature Mental Health, authors including former NIMH director Thomas Insel argue that psychiatry is being reshaped in two directions: economic pressures have pushed many psychiatrists towards brief medication management, while new procedural treatments increasingly require specialist technical skills and closer involvement in delivering care. They propose defining interventional psychiatry around combining psychiatric diagnosis and ongoing patient care with procedural expertise, supported by dedicated training and clearer standards. More broadly, they argue this could help rethink the psychiatrist’s role beyond prescribing medication. (link)
That’s all for this week.
Keep fighting the good fight!
Steve Duke
Founder of Hemingway
P.S. If you find these updates valuable, consider becoming a Hemingway Pro Member. As a Pro Member, you will get access to even deeper industry insights and will be invited to join our community of over 500 mental health leaders.
