A six-level clinical autonomy framework for artificial intelligence in dentistry

Artificial intelligence in dentistry is rapidly progressing from assistive decision support toward systems capable of executing clinical tasks with increasing autonomy. Despite these advances, the field lacks a structured framework to define, classify, and govern varying levels of clinical autonomy across diagnostic, procedural, and workflow domains in dental practice. This Perspective introduces a dentistry-specific six-level (L0-L5) conceptual clinical autonomy framework characterizing AI systems based on agentic capability, delegated decision authority, human oversight, clinical operating domain, and risk. The proposed taxonomy spans six levels (L0-L5), progressing from human-controlled systems (L0) through assistive (L1), advisory (L2), conditional (L3), and high-autonomy systems (L4), to full operational autonomy within defined clinical contexts (L5). A key inflection point is identified at Level 3, where systems transition from advisory outputs to delegated execution within defined clinical boundaries, marking a shift in responsibility, regulatory classification, and safety requirements. The framework emphasizes functional-level classification, recognizing that autonomy may vary across perception, decision-making, and execution components within hybrid systems. It integrates human-centered considerations, including clinician-AI interaction, transparency, interpretability, and evolving accountability models, while emphasizing inclusive validation and context-aware deployment across diverse patient populations and healthcare settings. By linking autonomy levels to proportional governance and staged translational evaluation, this conceptual framework is intended to support discussion of the safe and responsible integration of AI systems in oral healthcare. The framework has not undergone empirical validation or formal consensus development and should therefore be interpreted as a conceptual taxonomy intended to support future research, regulatory discussion, and refinement.