Table 1.

Recommendations for risk mitigation applying AI for suicide prevention in healthcare settings.

Domain	Recommendation for implementation	Recommendation for research
Consent	Develop informed consent for patients to sign detailing the actions and limitations of AI	Develop consent forms to all literacy levels and test for understanding
	Develop similar consent for providers	Develop patient education materials that detail the purpose of AI and evaluate for understanding
	Provide patients with “opt-out” of AI monitoring
	Provide time limits or expiration to consent
	Re-consent each year with evolving technology
	Have consent documents approved by experts and medical review board
Controls	Adopt standards for suicide monitoring with AI, such as determining what percentage of at-risk individuals will be monitored	Compare provider-informed vs. AI-only model to assess for increased accuracy with feedback
	Form an AI oversight panel with multidisciplinary specialty
	Request provider feedback routinely and update systems accordingly
	Create a system for providers to defer or activate risk monitoring with explanation
	Log model successes and failures, re-train models
Communication	Conduct focus groups with stakeholders to assess for appropriateness and utility of integrating AI into healthcare	Develop provider materials and elicit feedback for appropriateness
	Provide communication materials for provider use to discuss AI and the monitoring process