FDA grants de novo authorization for first autonomous AI colonoscopy triage system

The SKOUT Autonomous de novo is the clearest signal yet that the FDA is willing to authorize AI-enabled devices that operate without a required real-time clinician concurrence step for the specific finding category the model is cleared on. For most of the last five years, FDA-cleared clinical AI has been structured as “assistive” or “concurrent read” — the model flags something; the clinician makes the call. SKOUT Autonomous moves one step past that: for sub-threshold polyp characterizations in a screening colonoscopy workflow, the system’s triage output can be acted on directly without the gastroenterologist independently confirming the characterization in-procedure.

Three structural elements make this different from an incremental AI clearance. First, the autonomy carve-out is narrow and defined — the authorization covers only screening colonoscopies in high-volume centers, only polyps below a specific size-and-morphology threshold, and only characterization (hyperplastic vs. adenomatous), not detection or resection decisions. The agency is piloting the autonomy concept inside a confined operational space rather than opening the door broadly.

Second, the post-market registry requirement is substantive, not performative. Every center using the autonomous mode must participate in a registry that captures per-case characterization, pathology concordance, and 1-year outcomes. The registry is funded by the vendor but governed by an independent clinical consortium, with pre-specified stopping rules if real-world concordance deviates from the pre-market validation. That is a template for how the FDA may structure future autonomous-AI authorizations.

Third, the labeling language around indication boundaries will be the template for every other vendor trying to pursue similar carve-outs. Expect to see submissions from autonomous-mode cardiology AI, autonomous ophthalmology screening tools, and autonomous dermatology triage products following roughly the SKOUT template — narrow indication, defined threshold, mandatory registry, named stopping rules.

For gastroenterology operations specifically, the implication is immediate. Centers that have been running SKOUT in assistive mode can now, with institutional approval and workflow changes, shift to autonomous mode for the carved-out finding category. The labor-efficiency case is real: resect-and-discard protocols become operationally feasible without a per-polyp pathology spend if the AI characterization is treated as the diagnostic record.