Investors Bet $17M That Therapist in the Loop Wins

2026-04-02

Conversational AI Watch

The news that moves policy, portfolios, and patient safety.

By Jess Jessop | April 2, 2026 | Issue #4

▶︎ WATCH

🎧 QUICK LISTEN

🎧 DEEP DIVE

📄 READ ON WEB

Jess's Take

A million people a week are discussing suicide with ChatGPT. That is not a statistic you skim past.

That number comes from OpenAI itself. And now M13 and a group of serious health-tech investors just put $17 million behind Jimini Health, a company built on a single premise: every AI therapy interaction should be visible to a licensed clinician.

Sound familiar? It should. That is the Therapist in the Loop thesis, and the venture market just validated it with real dollars.

Meanwhile, the first randomized controlled trial of a generative AI therapy chatbot just landed in NEJM AI. A 51% reduction in depression symptoms. Published in one of the most respected journals in medicine. The clinical evidence era for AI therapy just began.

And in California, a judge is now consolidating a dozen mental health lawsuits against OpenAI into a single proceeding. The legal infrastructure for AI therapy accountability is being built in real time.

Here is what you need to know right now.

THE SIGNAL / MARKET AND INVESTMENT

Jimini Health Raises $17M to Build the Clinician-Supervised AI Therapy That Regulators Want

Jimini Health just closed a $17 million seed round led by M13, with Town Hall Ventures, LionBird, Zetta Venture Partners, and OneMind joining. Total raised to date: over $25 million.

Their product, Sage, is a therapy chatbot that operates between scheduled appointments, under direct clinician oversight. Every single conversation is visible to the treating provider. The AI follows care plans created by physicians and never improvises treatment.

The target market: 5.4 million Americans, many of them adolescents and young adults, who are already using ChatGPT and other general-purpose chatbots for mental health support without any clinical guardrails.

M13 partner Morgan Blumberg put it plainly: "When 1 million people a week are discussing suicide with a product that was never designed to handle it, that is not an edge use case."

Takeaway

For builders: The Jimini architecture mirrors what we are building with Casey AI. Clinician visibility into every conversation, care-plan-driven AI, real-time alerts. This is the design pattern that gets funded and gets through regulatory review.

For investors: M13 did not invest in another consumer chatbot. They invested in clinical infrastructure. That is the signal.

For clinicians: Sage is designed to extend your reach between sessions, not replace you. Every interaction stays in your line of sight. This is how AI earns clinical trust.

For Legislators

Jimini proves the market can build clinician-supervised AI therapy without waiting for regulation. But regulation still matters. Require clinical oversight as a condition of market access for any AI product that interacts with people in mental health contexts. The companies willing to build this way deserve a clearer path to market than the ones that are not.

Sources: SiliconANGLE | STAT News | MedCity News

THE SIGNAL / CLINICAL RESEARCH AND SAFETY

Landmark: First Randomized Trial Proves AI Therapy Chatbot Reduces Depression by 51%

This is a milestone for the entire field. NEJM AI just published the first randomized controlled trial of a fully generative AI therapy chatbot, and the results are significant.

Therabot, developed at Dartmouth, was tested on 210 adults with major depressive disorder, generalized anxiety disorder, or risk factors for eating disorders. The numbers: 51% reduction in depression symptoms. 31% reduction in anxiety. 19% reduction in eating disorder risk.

Participants sent an average of 260 messages over four weeks and rated their therapeutic alliance with the chatbot comparably to human outpatient therapy. That last finding alone should change how skeptics talk about AI-assisted care.

This is what peer-reviewed clinical evidence looks like. It is the model every serious AI therapy company should be pursuing. And it opens the door for the next generation of trials, including those that add clinician supervision to the protocol and measure whether outcomes get even better.

Takeaway

For builders: The RCT is now the benchmark. Therabot proved it can be done. The next step is trials that test clinician-supervised AI against standalone chatbots. That is where the Therapist in the Loop thesis gets validated with hard data.

For investors: Peer-reviewed clinical evidence de-risks AI therapy investments. Companies pursuing RCTs are building the regulatory and payer case that unlocks real revenue. This is the evidence standard that separates serious platforms from chatbot experiments.

For clinicians: A 51% reduction in depression symptoms is clinically meaningful. This is not replacing you. It is proving that AI-assisted care works, and the next question is how much better it works with you in the loop.

For Legislators

This study proves that rigorous clinical trials of AI therapy are possible and produce meaningful results. Encourage this. Create regulatory pathways that reward companies willing to do the hard work of clinical validation. The RCT model protects clients and gives policymakers the evidence base they need to make informed decisions.

Sources: NEJM AI | Dartmouth | Nature Mental Health

THE SIGNAL / LITIGATION AND LEGAL

OpenAI Mental Health Lawsuits Consolidated in California as Company Scrambles on Safety

A California court has consolidated 12 or more mental health lawsuits against OpenAI into a single judicial coordination proceeding. Plaintiffs' attorneys say more cases are coming.

The allegations are severe. The Social Media Victims Law Center claims OpenAI knowingly released GPT-4o despite internal warnings it was "dangerously sycophantic and psychologically manipulative." The suits allege the company compressed months of safety testing into a single week to beat Google's Gemini to market.

Claims include wrongful death, assisted suicide, involuntary manslaughter, and product liability. Multiple cases involve AI-induced psychosis and psychiatric hospitalization.

OpenAI's response: a new "trusted contact" feature that lets adult users designate someone who gets notified when they may need support. Plus $2 million in external mental health research grants. Whether courts view this as genuine safety improvement or litigation-driven damage control will matter.

Takeaway

For builders: Consolidation means coordinated discovery. Internal safety documents, testing timelines, and deployment decisions will be examined under oath. Build your safety record now, before litigation forces the question.

For investors: Litigation risk is no longer theoretical for AI therapy companies. Due diligence must include safety testing protocols, adverse event tracking, and clinical oversight architecture.

For clinicians: These cases are building the legal framework for AI therapy accountability. The standard of care for AI-assisted mental health treatment is being defined in courtrooms right now.

For Legislators

Do not wait for courts to define AI therapy safety standards case by case. Mandate adverse event reporting for any AI system used in mental health contexts. Require pre-deployment safety testing with published results. The allegation that a company compressed months of testing into one week should not be legally permissible.

Sources: Blockchain News | SMVLC | Law.com

THE SIGNAL / CLINICAL RESEARCH AND SAFETY

People Perceive AI Chatbots as More Judgmental Than Humans, Even with Identical Conversations

Researchers at UT Dallas, Temple University, and Indiana University ran four experiments with over 2,000 participants. Everyone saw the exact same mental health screening conversation. The only difference: half were told it was a human therapist. Half were told it was an AI chatbot.

The chatbot group rated the interaction as significantly more judgmental. They said the AI lacked deep understanding of what it means to experience mental health issues. One participant's summary: "The machine cannot convey those feelings."

The clinical implication is direct. People who feel judged are less likely to disclose symptoms, follow recommendations, or return for help. The same bias that keeps people away from human therapists also keeps them away from AI ones, just for different reasons.

Takeaway

For builders: Branding your product as "AI therapy" may actively reduce trust and disclosure. Consider how you frame the human-AI relationship. A clinician-supervised AI positioned as "your therapist's assistant" may outperform a standalone chatbot, even if the underlying model is identical.

For clinicians: This validates what many of you already know. The therapeutic relationship matters, and people do not form it with machines the same way. AI works best when it extends your relationship with your client, not when it tries to replace it.

For Legislators

Disclosure requirements matter, but they are not enough. Requiring AI chatbots to identify themselves as AI is necessary. But this study shows that disclosure alone reduces trust and willingness to engage. Policy must also ensure that human clinician oversight is part of the care model, not just a label on the screen.

Sources: UT Dallas News | MIS Quarterly

What We Built

Casey AI: The Therapist in the Loop Standard

Casey AI is our voice-first AI-native mental health EHR with Casey Life and Peer AI Coach supervised by licensed therapists. 1.78 million safety tests passed. 100% accuracy. Every AI interaction visible to the supervising clinician. Pre-FDA safety validation complete, pursuing 510(k) clearance.

The client talks to Casey. Casey talks to the therapist. The therapist stays in the loop. That is the architecture that earns clinical trust, regulatory approval, and investor confidence.

Watch the Demo →