AI·SPB publishes a concept for an experimental legal regime covering narrowly defined medical AI scenarios. Its page identifies working version 0.3, dated 19 August 2026, and states that the EPR has not launched and is not an official project of the Government of Saint Petersburg. The concept lists ten distinct clinical candidates and a 48-month evaluation path across stages A0–A4, from an assistive prompt to a potentially legally significant action. These are scenarios for assessment, not a list of autonomous services already permitted. Each use would separately require evidence, a registered product, an operator, a clinical site and a legal pathway. Uncertainty is routed to a human. The practical value is a structured discussion of permission by task, system version and control conditions, with refusal, escalation, measurable outcomes and stop criteria defined in advance. The publication describes a concept; it does not itself authorize autonomous action. Reduced time, less harm or improved outcomes are aims to evaluate, not confirmed effects. Every scenario needs its own legal, clinical and technical assessment.
Project review
AI·SPB: assessing autonomy scenario by scenario
AI·SPB publishes a concept for an experimental legal regime covering narrowly defined medical AI scenarios. Its page identifies working version 0.3, dated 19 August 2026, and states that the EPR has not launched and is not an official project of the Government of Saint Petersburg. The concept lists ten distinct clinical candidates and a 48-month evaluation path across stages A0–A4, from an assistive prompt to a potentially legally significant action. These are scenarios for assessment, not a list of autonomous services already permitted. Each use would separately require evidence, a registered product, an operator, a clinical site and a legal pathway. Uncertainty is routed to a human. The practical value is a structured discussion of permission by task, system version and control conditions, with refusal, escalation, measurable outcomes and stop criteria defined in advance. The publication describes a concept; it does not itself authorize autonomous action. Reduced time, less harm or improved outcomes are aims to evaluate, not confirmed effects. Every scenario needs its own legal, clinical and technical assessment.