The evolution of surgical robotics is increasingly driven by artificial intelligence supporting perception, decision-making, and task execution. As AI-enabled capabilities introduce new forms of automation into surgical systems, patient safety depends not on automation itself, but on how these capabilities are bounded, supervised, and governed. This presentation presents a visionary yet practical perspective on fail-safe by design as a foundation for trustworthy, risk-aware autonomy in surgical robotics.

Using the Polaris surgical robotic system as a case study, we describe how risk management was embedded from early development to support safe automation in high-precision procedures, including eye surgery. Polaris enables AI-assisted task execution under continuous surgeon supervision. System-level risk analysis informed architectural decisions, autonomy boundaries, and human–machine interaction, particularly for AI components subject to uncertainty and variability.

In this framework, fail-safe behavior for task-level autonomy is not a single static response. Instead, it is dynamic and context-driven. The appropriate fail-safe action depends on clinical context, including the current surgical step, anatomical conditions, tissue state, and proximity to critical structures. A failure occurring during approach, for example, requires a different mitigation than the same failure during tissue engagement. Fail-safe behavior is therefore defined as a controlled transition to the safest achievable state for the given context, minimizing harm while preserving tissue integrity and restoring surgeon authority.