The conversation about AI in mental health has moved beyond hype to a harder question: what has to be true before these tools can help at scale? In a new paper, Martin Paulus and John Torous, with colleagues, argue that the main barriers are no longer computational but infrastructural (Paulus et al., 2026). For South African practitioners weighing these tools, their five priorities offer a clear-eyed roadmap.
Consequently, the paper is less about clever algorithms and more about the systems around them. Below, we summarise the five priorities and what they mean for responsible practice.
Why infrastructure, not algorithms, is the barrier
The authors synthesise insights from a 2026 study group of the American College of Neuropsychopharmacology (Paulus et al., 2026). Their central claim is striking: AI has entered psychiatry at scale, yet its clinical impact remains limited by a gap between technical validation and real-world use.
Specifically, the obstacles are unreliable measurement systems, incomplete governance, and weak standards for clinical evidence. Therefore, better models alone will not deliver better care. The surrounding infrastructure has to catch up first.
The five priorities for AI in mental health
The paper sets out five priorities required for genuine clinical impact. Together, they define a practical agenda rather than a wish list (Paulus et al., 2026).
- Robust measurement and phenotyping infrastructure, including reliable psychometrics and standardised data pipelines.
- Clinician-facing augmentation tools that reduce workflow burden, such as ambient documentation.
- Patient-facing tools that are rigorously tested before wider rollout.
- Equity-focused governance that builds fairness into oversight.
- A shift toward causal models, not merely predictive ones.
Moreover, the ordering is deliberate. The authors see the most immediate, scalable benefit in tools that support clinicians, rather than in tools that interact directly with patients.
Clinician-facing tools come first
According to the paper, clinician-facing augmentation offers the most immediate value (Paulus et al., 2026). Ambient documentation, for example, can reduce administrative burden and free clinicians to focus on the person in front of them.
This emphasis is reassuring for practitioners wary of AI. Specifically, it positions the technology as a support to your work, not a substitute for it. Furthermore, tools that reduce burnout-inducing paperwork address a real and pressing problem.
Understanding these developments is part of staying ethically and professionally current.
Patient-facing tools, equity, and causal models
For patient-facing tools, the authors urge caution and rigorous testing before any wide release. Consequently, chatbots and self-guided applications should meet a high evidence bar, given the vulnerability of the people they serve.
Equity-focused governance forms the fourth priority. In particular, oversight structures should be designed so that fairness is built in from the start, rather than added as an afterthought. This concern is especially relevant in South Africa, where access to care is deeply unequal.
Finally, the authors call for a shift from predictive to causal models. Predictive models can flag associations, but causal models help explain why something happens. Therefore, they are more likely to support sound clinical decisions.
Safety, equity, and effectiveness
Across all five priorities runs a single theme. Success will come not through model performance alone, but through disciplined integration into workflows, measurement systems, and governance that ensure safety, equity, and real-world effectiveness (Paulus et al., 2026).
What this means for South African mental health care
For local practitioners, the roadmap is a useful filter. When you encounter a new AI tool, you can ask whether it rests on reliable measurement, whether it supports rather than replaces your judgement, and whether it has been tested fairly.
Moreover, the equity focus resonates strongly here. Any responsible rollout of AI in mental health must widen access rather than deepen existing gaps. These priorities also complement the professional ethics set out in recent guidance on AI in clinical practice.
Staying informed is part of ethical, forward-looking practice. To keep building your knowledge while earning CEUs, browse our 2026 CPD packages for South African psychology professionals.
References
- Paulus, M. P., Torous, J., Perlis, R. H., Sarma, K. V., Ajilore, O., Khalsa, S. S., & Rodriguez, C. I. (2026). Responsible and innovative AI for mental health care: Five priority themes. NPP—Digital Psychiatry and Neuroscience, 4(1), 17. Retrieved from https://doi.org/10.1038/s44277-026-00068-x
- Linardon, J., Firth, J., Carvalho, A. F., et al. (2026). Multidisciplinary research priorities for artificial intelligence in mental health: A call to action. The Lancet Psychiatry. Retrieved from https://doi.org/10.1016/S2215-0366(26)00127-6
- DeCamp, M., Sulmasy, L. S., Karches, K. E., et al. (2026). Ethics and professionalism in artificial intelligence and medical practice: A position paper from the American College of Physicians. Annals of Internal Medicine. Retrieved from https://doi.org/10.7326/ANNALS-26-01792