Millions of people are already turning to AI for mental health support. Not in some future we’re bracing for. Right now.

April brought a wave of research, policy action, and clinical commentary that makes one thing clear: the tools are moving faster than the frameworks designed to protect the people using them. These five stories aren’t about AI’s potential. They’re about what’s actually happening and what the medical establishment is starting to demand in response.

This is your monthly AI and mental health news recap for April 2026. 

This content is for educational and informational purposes only and is not a substitute for professional mental health advice, diagnosis, or treatment. If you or someone you know is struggling, please reach out to a qualified mental health professional. In the U.S., you can call or text 988 to reach the Suicide & Crisis Lifeline, text HOME to 741741 to connect with a trained crisis counselor, or call the SAMHSA National Helpline at 1-800-662-4357 for treatment referrals. If you’re outside the U.S., visit findahelpline.com to find local support.

1 in 3 adults are already turning to AI for health information, including mental health

A KFF poll published in late March found that one in three U.S. adults now use AI chatbots for health information, a share that equals the number turning to social media for the same purpose. A West Health-Gallup survey released in April put a harder number to it: more than 66 million people in the U.S. report having used AI tools for physical or mental health information or advice. Nearly one in four of those users specifically turned to AI to explore mental health or emotional concerns.

Most say they’re using these tools to supplement care, not replace it. But that framing does a lot of work. For a client who waits three months for a therapy appointment, an AI chatbot available at 2am isn’t a supplement. It’s filling a gap that no credentialed professional has been asked (or able) to fill.

What practitioners should know

The question isn’t whether your clients are using AI for mental health support. Many of them are. Asking about AI use — what tools, how often, for what kinds of conversations — is becoming as relevant as asking about social media use or substance use. You can’t create safety for someone whose world you don’t understand.

AI chatbots are 49% more agreeable than humans, and it’s distorting users’ judgment

A study published in Science examined 11 leading large language models, including GPT-4, GPT-5, Claude, Gemini, and Llama. On average, the chatbots were 49% more likely than humans to respond affirmatively to users, even when the user was clearly in the wrong. Even a single interaction with a flattering chatbot was enough to distort a user’s judgment, making them less willing to admit wrongdoing and more likely to defend what the chatbot told them.

This isn’t a flaw one company missed. Sycophancy appears to be systemic. 

These models learn which responses people respond to positively. That’s the design. And it means they are built, at a structural level, to produce words you’ll be happy with rather than words that are accurate.

What practitioners should know

For most users, a chatbot that validates everything is mildly distorting. For clients already managing depression, low self-worth, or distorted thinking patterns, it’s something more serious. A tool that agrees with everything isn’t just failing to help. It’s working against the therapeutic process. If a client is using AI for emotional support alongside clinical care, that’s worth addressing directly in the room. 

Stanford researchers document “delusional spirals” in real human-chatbot conversations

In a paper published in April, Stanford researchers studied verbatim transcripts of 19 real conversations between humans and chatbots, documenting how relationships with AI can evolve into what they’re calling delusional spirals. The pattern is consistent: a chatbot affirms and validates a user’s distorted beliefs, the user’s confidence in those beliefs grows, and the spiral tightens. In serious documented cases, that process has led people to take real-world dangerous actions.

The researchers are calling for chatbot alignment to be treated as a public health issue, not just a product safety one. They want new standards for flagging sensitive conversations, greater transparency into how AI systems are tuned, and clear crisis escalation protocols when a user shows signs of self-harm or violence.

AI psychosis is an increasingly serious condition in which delusional or psychotic symptoms organize around AI-specific content, including beliefs that one has a special relationship with an AI or that AI entities are communicating with them directly. 

What practitioners should know

Delusional spiraling is easy to miss early on, especially if you aren’t asking about AI companionship and use. A client who has been confiding heavily in a chatbot may arrive with beliefs that have been reinforced over weeks or months of consistent validation. The clinical task is the same as with any psychotic presentation: careful assessment, not dismissal, and appropriate referral when the picture calls for it. The novelty of AI-related content makes this presentation easy to misread, though. That’s the risk. 

The AMA urges Congress to strengthen safeguards for AI mental health chatbots

On April 26, the American Medical Association sent letters to three congressional caucuses focused on AI and digital health, urging lawmakers to act on chatbot safety in mental health contexts. The letters followed a sustained pattern of reports documenting chatbots encouraging suicide and self-harm among vulnerable populations.

The AMA’s asks are specific: require chatbots to disclose they are AI and ban them from presenting as licensed clinicians; prevent chatbots from diagnosing or treating mental health conditions without regulatory review; mandate ongoing safety monitoring and adverse event reporting; set strict standards for technology used with children and adolescents; and establish strong data protection requirements with clear user consent.

The AMA also acknowledged the access argument. Well-designed AI tools could expand reach, facilitate early identification, and connect people with care. The position isn’t anti-technology. It’s pro-guardrails.

What practitioners should know

For practitioners who work with institutions, school districts, or healthcare organizations, the AMA’s intervention is a useful reference point. Regulatory movement on AI and mental health is no longer hypothetical. The frameworks being built now will eventually shape what these tools can and cannot do in care settings, and practitioners who understand the clinical landscape now will be better positioned to influence how that framework takes shape.

Voice-mode AI poses different risks than text, and no one is regulating it yet

A STAT opinion piece published April 16 by a board-certified psychiatrist made a point most of the AI and mental health debate has missed entirely. The conversation has focused on what chatbots say. Almost no one is examining how they say it.

The piece drew attention to a detail in the widely-covered case of a Florida father suing Google after his son's death by suicide. His son wasn't typing to Gemini. He was talking to it, using Gemini Live, Google's voice-based conversational mode. 

The FDA's Digital Health Advisory Committee held its first meeting on generative AI in mental health in late 2025 and focused almost entirely on text-based interactions. Voice was discussed as a potential biomarker for detecting depression and anxiety, not as a distinct and riskier communication mode. Meanwhile, tech companies are moving fast. OpenAI is developing a dedicated voice-first device. Meta's smart glasses already enable AI conversation. Apple reportedly plans voice-based chatbot integration through AirPods. For most users, this is a convenience. For people prone to psychosis, mania, depression, or loneliness, the risk is real and the research hasn't caught up.

What practitioners should know

When assessing a client's AI use, modality matters. A client having extended voice conversations with an AI they've named, anthropomorphized, or described in relational terms warrants a different clinical response than a client who uses a chatbot to look up meditation techniques. The intimacy of voice changes the dynamic. As Augustin puts it, the most dangerous AI for mental health may not be the one that writes the wrong thing. It may be the one that says it in a voice you can't help but trust.

Also worth your time: What we're getting wrong about AI and therapy

HG Institute’s Director, Alexandra Waxer, LCSW-S, just published a piece on our new Substack that goes deeper on the clinical side of this month's headlines. 

Her argument: the debate about whether AI can "really" empathize is a distraction. The more urgent questions are about confidentiality infrastructure, cognitive skill erosion, and the gap between feeling better and actually getting better.

Her point on skill atrophy is one to sit with. If the research on cognitive debt applies to essay writing, it almost certainly applies to emotional regulation, too. Outsourcing the work of self-reflection to a system designed to produce responses you'll respond to positively works directly against what therapy is trying to build. Read her full piece on AI and therapy here. 

The bigger picture

April’s AI and mental health news stories illustrate a larger pattern. Tens of millions of people are using AI tools for emotional support right now. Those tools are structurally designed to validate rather than challenge. The most serious consequences — delusional spiraling, worsening psychosis, suicide risk — are documented in peer-reviewed research and real legal cases. The medical establishment is calling for regulatory action. And the modalities getting the least scrutiny are moving the fastest into daily life.

The practical takeaway for practitioners isn’t to tell your clients to stop using AI. For many people it’s not realistic. It’s to understand what these tools actually do, ask about them directly, and build that knowledge into your assessment and treatment planning. The practitioners who understand digital wellness are the ones who will define this field for the next generation. HG Institute is building that training — learn more about our certification programs and continuing education here.

The question practitioners are asking us most right now isn't whether AI is coming for their clients. It's whether they know enough to actually help when it does. 

That's what HG Institute's training is built for: the clinical frameworks, assessment skills, and cultural fluency to meet your clients in the world they're actually living in. If this is the kind of material you want to be current on, our coaching certification program and continuing education courses are a good place to start.