2026-06-02A recurring clinical intuition holds that if patients could only see the faulty machinery of their own thinking — the distortions, the dissonance, the leaps of inference — they would be freed from some part of their suffering. This whitepaper takes that intuition seriously and then takes it apart. Four clinical moves are routinely collapsed into one: cognitive restructuring, dissonance/discrepancy work, metacognitive bias training, and formal fallacy instruction. The first three rest on distinct and substantial literatures; the fourth is mostly an open research question. The evidence supports surfacing in specific forms — strongest when it is collaborative rather than didactic, self-generated rather than imposed, and depersonalized rather than confrontational — and gives little support for declarative correction from authority.
2025-11-08For many people, primary care has become the first door they knock on—and sometimes the only one—when they are struggling with depression, anxiety, insomnia, the aftermath of trauma, difficulty concentrating, or substance use. Yet specialty mental health care remains hard to reach, especially in rural and underserved communities, which leaves primary care clinicians to meet needs that can outstrip their training. Artificial intelligence (AI) can help these clinicians recognize common, lower-acuity conditions and choose evidence-based first- line care—but only when it serves as clinical decision support rather than an autonomous system that diagnoses or prescribes on its own. This paper proposes a safety-bounded model for AI-aided assessment of “uncomplicated” mental illness in primary care, built on validated
2025-03-02Patient-facing large language models are entering behavioral health, whether the field is ready or not. Some systems are purpose-built for mental health. Others are general commercial AI platforms that become behavioral health tools because users bring them behavioral health questions. Either way, the safety problem is no longer theoretical.
