New review maps how algorithms are being used to assess risk, spot mental health needs, and guide care in the justice system—while warning about bias, privacy, and “black‑box” decisions.

A narrative review led by David B. Olawade, PhD, finds that artificial intelligence (AI) is moving quickly from research labs into forensic mental health—the branch of care that intersects with courts, prisons, and secure hospitals. The authors describe fast‑growing uses of AI for risk assessment, screening and diagnosis, behavioral analysis (from speech to facial expressions), and treatment planning, with the potential to make high‑stakes decisions more consistent and data‑informed.

What AI can do now

• Predict risk more consistently: Tools trained on large datasets (including widely discussed systems like COMPAS) can estimate the likelihood of re‑offense or harm with accuracy around 70–75% in some studies—often outperforming ad‑hoc judgment when properly calibrated.
• Spot problems earlier: Algorithms that analyze language, phone use patterns, or interview recordings can flag signs of depression, anxiety, agitation, or deception sooner than traditional methods, enabling earlier intervention.
• Support overburdened systems: By triaging cases and standardizing parts of evaluations, AI can help allocate scarce mental‑health resources where they’re needed most.

The risks (and why they matter)

• Bias baked into data: If historical records are incomplete or skewed, AI can entrench disparities—especially along racial and socioeconomic lines.
• Opaque decisions: Many models operate as “black boxes,” making it hard for judges, clinicians, or defendants to understand why a score was assigned.
• Privacy and consent: Sensitive health and justice data raise serious concerns about who sees what—and how information can be reused.
• Legal accountability: When an algorithm gets it wrong, responsibility is unclear: the developer, the clinician, or the institution?

What the authors recommend

• Better data, shared responsibly: Build representative, ethically sourced datasets (including open‑source efforts with safeguards) to improve reliability.
• Human‑in‑the‑loop by design: Use AI to complement—not replace—expert clinical judgment, with ongoing validation to cut false positives/negatives.
• Explainable AI (XAI): Favor models that show how key factors drive their outputs so decisions can be challenged and improved.
• Harmonized rules: Align oversight across jurisdictions. The review highlights existing frameworks like HIPAAand FTC authority in the U.S., GDPR in Europe, and emerging efforts such as the EU AI Act, plus guidance in Canada and Australia—while noting gaps for forensic settings.
• Independent testing and audits: Mandate regular bias, performance, and security checks, with public reporting where possible.

Bottom line for readers:
AI could make forensic mental health assessments fairer and more timely—but only if we pair innovation with transparency, independent oversight, and strong privacy protections. Until then, the safest path is AI with humans, not AI instead of humans.

Source: Olawade DB, Ayoola FI, Ebo TO, Asaolu AJ, Egbon E, David‑Olawade AC. “Artificial intelligence in forensic mental health: A review of applications and implications.” Journal of Forensic and Legal Medicine. 2025;113:102895. DOI: 10.1016/j.jflm.2025.102895.

Editor’s note: This article is for information only and is not a substitute for professional medical advice.