May 2026 brought a wave of new research and policy action that collectively make one thing harder to ignore: the gap between how fast AI mental health tools are spreading and how equipped practitioners are to respond to this transformation. Three peer-reviewed studies examined how people are actually using AI tools and whether they help. A new Oregon law set a new precedent for crisis safety requirements on AI companion platforms. And public concern about AI's impact on mental health reached a new high.

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

Americans' concern about AI and mental health has reached a new high

A May 2026 YouGov survey of 1,000 adults found that 43% of Americans say they are very concerned about the possibility of AI making mental health problems worse, up from 35% in June 2025. The increases were sharp across age groups: among adults 45 to 64, concern rose from 37% to 50%; among adults under 30, from 29% to 45%.

Two-thirds of Americans say they would be uncomfortable working with an AI therapist. Yet adults under 30 are about twice as likely as older Americans to say they would be comfortable with one (37% vs. 20%). One in ten Americans say they could probably or definitely form a deep emotional bond with an AI chatbot. One percent say they already have.

These numbers tell two stories simultaneously. 

Public skepticism about AI in mental health is growing across age groups. And the population most likely to encounter AI mental health tools — younger people, with higher exposure and greater comfort — is also the population showing the sharpest rise in concern. They're not naive about this. They're living in it.

What practitioners should know

Pay attention to the generational split in these numbers. Your younger clients may be both more likely to be using AI for emotional support and more likely to have mixed or complicated feelings about it. That ambivalence is worth exploring directly rather than assuming comfort or discomfort. The same client who uses a chatbot every day may also be genuinely uncertain about whether it's helping them.

Brown University documents systematic ethics violations in AI therapy sessions

A study out of Brown University, covered by Psychology Today, found that large language models used in therapeutic contexts routinely violate the professional standards that govern human mental health care. Researchers worked with seven trained peer counselors and three licensed psychologists to evaluate 137 sessions across multiple LLM models.

The violations fell into five categories: 

• Lack of contextual understanding

• Poor therapeutic collaboration

• Deceptive empathy

• Unfair discrimination

• Inadequate crisis management

The researchers documented violations across every major dimension of ethical therapeutic practice. Models minimized clients' identity-based experiences, dominated conversations in ways the research team described as gaslighting, and displayed empathy that looked real but lacked genuine understanding. They showed bias against non-dominant identities and — most critically — fell consistently short when clients disclosed trauma, abuse, or suicidal ideation.

The researchers were clear about what this means: psychotherapy is a relational and ethical practice, not a language generation task. Treating it as the latter carries real clinical risk.

What practitioners should know

If clients are using any of these LLM models (or apps built on them) for therapeutic conversations, you're not starting from a neutral baseline. You may need to actively work against patterns those tools have reinforced. Asking directly about AI use, including what the tool said and how the client responded to it, belongs in your standard assessment. Not because AI use is inherently harmful, but because these interactions aren't necessarily clinically neutral.

Drexel University maps real-world AI mental health use through 4 million Reddit posts

A Drexel University study, which will be presented at the 2026 Annual Meeting of the Association for Computational Linguistics, analyzed over 4 million posts across 47 mental health subreddits to understand how people actually use AI chatbots for emotional and mental health support. The research team, led by Shadi Rezapour, PhD, focused on a sample of 5,126 posts.

Most users approached AI as a supplement to human therapy, turning to it for emotional reassurance, coping strategies, and practical guidance during moments when professional care was unavailable, inaccessible, or insufficient. Many also used it specifically for ADHD- and autism-related support tasks like organization and structuring their days.

The study also identified what researchers called a "bond paradox." 

When AI helped users accomplish specific goals — reflection, coping, practical problem-solving — experiences tended to be positive. But when a strong emotional bond formed without those clear task-based anchors, especially in companionship use cases or repeated reassurance-seeking, the outcomes were more likely to include emotional dependence, worsening symptoms, shame, and difficulty disengaging. Roughly 51% of posts mentioning AI mental health use also mentioned concerns about its risks and limitations.

What practitioners should know

Your clients who use AI are mostly not trying to replace you. They're filling gaps between sessions, at 2am, when the cost of care is too high, or your waitlist is too long. Understanding the specific use case matters clinically. Task-focused AI use looks very different from companionship-seeking, and the clinical picture for each is meaningfully different. Worth asking: what are they actually doing with it, and what does the relationship feel like to them?

BMC Geriatrics meta-analysis finds AI agents reduce depression in older adults, but don't touch loneliness

A systematic review and meta-analysis published in BMC Geriatrics examined eight randomized controlled trials covering 611 older adults to assess whether AI-based conversational and socially assistive agents actually work for depression and loneliness. The findings were precise and worth sitting with.

For depressive symptoms, the evidence was statistically significant: AI-based interventions showed a small but meaningful effect (Hedges' g = −0.25). Cognitive-focused approaches produced more consistent results than companionship-focused ones. For loneliness, there was no significant effect at all, and the heterogeneity across studies was substantial enough that no reliable conclusion could be drawn.

The authors concluded that AI agents may serve as a useful adjunct for depression support in older adults — particularly for those with barriers to in-person care — but the evidence does not support their use as a loneliness intervention.

What practitioners should know

For practitioners working with older adults, this is one of the cleaner studies we have on what AI actually does and doesn't do in a specific population. The cognitive and task-focused finding maps directly onto the Drexel bond paradox finding above: structure and goals seem to matter more than relational warmth. If you're advising clients or institutions on AI companion tools for this population, the evidence supports a narrow and well-defined role and argues against positioning these tools as a loneliness solution, which is often how they're marketed.

ChatGPT adds a "Trusted Contact" feature for crisis situations

OpenAI announced that ChatGPT would allow users to designate a trusted contact who can be notified if the AI detects signs of a mental health crisis. The announcement, shared on LinkedIn by researcher Declan Grabb, represents one of the more concrete steps any major AI company has taken toward integrating human-in-the-loop crisis response. 

The design acknowledges something the research has been pointing toward for months: that a person in crisis reaching out to AI may need a pathway to a real person, and that the AI itself can't provide what that moment actually requires.

What practitioners should know

This is a feature worth knowing about because your clients may not know it exists or may not have set it up. If you're working with clients who are using AI companionship tools or chatbots for emotional support, it's worth asking whether they've designated a trusted contact and who that person is. It's also a useful conversation starter about what AI can and can't do in a moment of genuine distress.

A digital therapy app outperforms campus counseling referrals in a large-scale study

A study published in Nature Human Behaviour followed more than 6,200 university students identified through campus-wide screening as being at high risk for or diagnosed with depression, anxiety, or an eating disorder. Students offered a smartphone app delivering CBT content alongside text coaching from human coaches showed better outcomes at six weeks, six months, and two years compared with students who received a standard referral to campus mental health services. Nearly 75% of students offered the app used it at least once. Only 30% of students who received a referral accessed any treatment in the following six months.

The study makes one design choice explicit: the app does not run on generative AI. The researchers drew a direct line between that choice and the study's credibility, noting that generative AI-based therapy remains largely untested and citing the American Psychological Association's November 2025 advisory against using generative AI chatbots as a replacement for standard mental health care.

What practitioners should know

The access gap the study documents is worth holding onto. When nearly half of students who complete initial screening show signs of depression, anxiety, or an eating disorder, and when only 30% of those referred to care actually receive any treatment, something is broken in the referral pipeline, not just the care capacity. Digital tools with clear CBT frameworks and human coaching components are showing some promise as a bridge. 

Oregon passes a law requiring AI companion apps to connect teens in crisis to human care

Oregon Governor Tina Kotek held a ceremonial signing for Senate Bill 1546 on May 7, making Oregon one of a growing number of states (alongside New York and California) to require crisis safeguards on AI companion platforms.

The law was championed by Lines for Life and a broad coalition of behavioral health advocates. Under the new requirements, platforms must detect language indicating suicidal ideation or self-harm, immediately connect users in crisis to the 988 Suicide & Crisis Lifeline or equivalent resources, apply additional protections when the user is a minor, and report crisis referral data to the Oregon Health Authority.

That last requirement matters as much as the safety features themselves. It creates a public record of whether platforms are actually connecting people to help, not just whether the feature exists somewhere in the code.

The context behind the legislation: research cited during the advocacy process found that three in four teens have used AI companion apps at least once, with half using them routinely. Until now, those platforms faced no requirement to respond when a user expressed thoughts of suicide or self-harm. A teenager could describe a crisis in detail and receive a response with no mention of help.

What practitioners should know

For practitioners who work with adolescents or within institutional settings, this legislative shift signals what's coming for the regulatory landscape more broadly. Oregon's law is already part of a national movement. The practitioners who understand these developments, and can explain them to clients, families, and school or healthcare institutions, are ahead of where most of the field currently is. Knowing what platforms are required to do in your state, and where those requirements still don't exist, is now a relevant piece of clinical context.

The bigger picture 

Across three different study designs and populations from AI and mental health news from May, a pattern holds: AI tools that help people accomplish specific, bounded tasks tend to produce some benefit. 

We’re starting to get a clearer picture of where AI adds value and where it doesn’t. However, there are open questions that still need answers: 

• What happens longitudinally? Most studies follow clients for months, not years. We don't have solid data on what repeated long-term AI emotional support does to the capacity for human connection or to the willingness to seek professional care when it's genuinely needed.

• What does ethical AI in mental health actually require? The Brown University study lays out a framework for the violations that currently exist. The APA has issued advisories. Oregon has passed legislation. But no comprehensive clinical standard has emerged for what an ethically compliant AI mental health tool should do in practice.

• How should practitioners assess and document AI use? There's no consensus yet on how to integrate AI use into clinical assessment, treatment planning, or documentation. As these tools become more prevalent, that gap will matter more.

Stay current with digital wellness training

The research coming out every month on AI and mental health is clinically relevant right now. Your clients are using these tools. The institutions you work with are evaluating them. The regulatory landscape is shifting. Practitioners who understand this space aren't just staying informed; they're building the expertise that defines what this field looks like for the next generation.

HG Institute's certification programs and continuing education courses are built for exactly this moment, giving you the frameworks, evidence base, and cultural fluency to work with the populations who need it most.