Objective:

To examine the regulatory gap surrounding LLM-associated clinical errors and propose a structured framework for legal liability in the context of generative AI in clinical settings.

Approach:

• Regulatory Analysis: The article analyzes current regulatory instruments, including the US FDA CDSS guidance and the EU Artificial Intelligence Act, in relation to generative AI.
• Hypothesis Development: It introduces the autonomy–liability correspondence hypothesis, suggesting that legal liability should correlate with the system's autonomy from human review.
• Framework Proposal: The article proposes a Three-Tiered Liability Escalation Framework (T-LEF) and Clinical Algorithmic Audit Trails (CAAT) to address accountability in AI systems.

Key Findings:

• Current regulatory frameworks inadequately address the non-deterministic outputs of generative AI.
• Automation bias can lead to over-reliance on AI-generated recommendations, increasing the risk of clinical errors.
• The proposed T-LEF allocates liability based on the degree of AI autonomy from human oversight.

Interpretation:

The article argues that existing medico-legal frameworks are misaligned with the operational characteristics of generative AI, necessitating new approaches to liability and accountability.

Limitations:

• The T-LEF and CAAT are prospective proposals requiring empirical validation and stakeholder consultation.
• The framework does not provide a finalized implementation blueprint.

Conclusion:

The article offers a structured contribution to regulatory deliberation regarding the integration of generative AI in clinical practice, particularly in pediatric care.