On June 4, 2026, the FDA cleared Sonio Suspect, a software product from the company Sonio, through the 510(k) pathway as clearance number K261519. The decision was a Special 510(k) — a streamlined submission type — received on May 7, 2026 and found Substantially Equivalent in under a month. The device sits in the Radiology advisory panel under product code POK, regulation 892.2060, as a Class II device.
Sonio operates in clinical-imaging AI, and 'Suspect' is exactly what its name implies: software that flags and helps characterize imaging findings that may be suspicious for cancer. As with every clinical-AI clearance, the interesting question is not whether the software is 'AI' but what role the FDA permits it to play. The agency answers that through the product code, and the POK classification defines the device type's function in one verbatim line.
"Assist users in characterizing lesions identified on acquired medical images"— FDA Product Classification, product code POK, source
The operative verb is assist, and the operative object is characterizing lesions that have already been identified. This is computer-assisted diagnosis — CADx — and the language is doing careful work. The software does not autonomously detect lesions, and it does not render a diagnosis. It helps a clinician characterize findings the clinician is already looking at.
Why 'characterize' is not 'diagnose'
In imaging-AI regulation, the distinction between detection, characterization, and diagnosis is everything. A computer-aided detection tool points to where something might be. A computer-aided characterization tool — what POK covers — helps describe the properties of a finding a user has identified, supporting the clinician's own interpretation. Neither is a substitute for the physician's diagnostic judgment. The POK definition's phrasing, 'assist users in characterizing,' places Sonio Suspect firmly in that assistive, second-reader posture.
That bounded role is the entire design of the regulatory category, and it is what allows a clinical-AI product to clear as a Class II device rather than facing the far higher bar of a standalone diagnostic. For strategists and IP analysts watching the imaging-AI landscape, the lesson repeats across vendors: the cleared claim almost always reads as 'assists' or 'aids,' and the marketing that implies autonomous diagnosis runs ahead of what the clearance supports. Scope ends where the product code's function statement ends.
The Special 510(k) tells its own story
Sonio cleared Suspect via a Special 510(k), and the route is itself informative. A Special 510(k) is reserved for modifications to a manufacturer's own legally marketed device where the change does not alter the intended use or the fundamental scientific technology, and where the resulting performance can be verified against established methods. Choosing this path — and clearing it in under a month — strongly suggests Suspect is an iteration on Sonio's existing cleared software rather than a new-from-scratch device. The intended use and core technology stayed within the bounds the company had already established.
That has practical implications. A Special 510(k) implies continuity: the characterization function, the imaging modality, and the assistive role are consistent with Sonio's predicate. The modification might be a model update, an expanded set of findings, or a workflow change — but whatever it is, it did not push the device outside the POK box. For a company building a clinical-AI portfolio, stacking Special 510(k)s on a stable predicate is the efficient way to ship improvements without re-litigating intended use each time.
The detection-vs-characterization line, in practice
To see why the POK framing matters, it helps to trace how a CADx tool fits into a real reading workflow. A clinician acquires an image and identifies a finding — a lesion the human eye has already flagged. Only then does the characterization software engage, offering an assessment of that finding's properties to inform the clinician's interpretation. The software never operates upstream of the human, scanning for things no one has noticed, and it never operates downstream as the final word. It sits in the middle, as a second opinion the clinician can weigh. That middle position is precisely what the verb 'assist' and the participle 'identified' in the POK definition encode.
This is the structural reason imaging-AI vendors converge on the same modest-sounding claims. The aggressive claim — autonomous detection or diagnosis — would move the device into a different and far more demanding regulatory posture, where the evidentiary bar and the liability exposure both rise sharply. The assistive claim keeps the device in Class II, clearable by 510(k), and tethered to a human decision-maker. For IP and product strategists, the cleared scope is therefore the most honest summary of what a clinical-AI product can defensibly do: less than the marketing implies, but exactly what the regulation describes.
What stacking Special 510(k)s reveals about the roadmap
The choice of a Special 510(k) for Suspect is also a window into Sonio's development cadence. Because the Special pathway requires that intended use and fundamental technology stay constant, a company that clears successive Special 510(k)s is signaling an iterative product strategy: it establishes a predicate once, then ships verified improvements against it without reopening the core regulatory questions each time. That is an efficient way to compound an AI product — model refreshes, expanded finding sets, workflow refinements — while keeping every release anchored to the same cleared intended use. A sub-month review here is consistent with that disciplined, modification-driven approach.
What the record establishes
The openFDA entry fixes, verbatim, that Sonio Suspect cleared June 4, 2026 as K261519, that it was a Special 510(k) found Substantially Equivalent, and that it is Class II software in the Radiology panel under product code POK. The classification record supplies the function statement quoted above. What the API entry does not yet expose is the named predicate device, the specific imaging modality, the lesion types, and the standalone or reader-study performance data — those details live in the full 510(k) summary.
The durable takeaway is the framing. Sonio Suspect is cleared to assist users in characterizing lesions suspicious for cancer — not to find them unprompted and not to diagnose them. In a market where 'cancer-detecting AI' makes for a better headline than the regulation allows, the cleared scope is the corrective. The software is a tool in the clinician's hands, and the FDA's product code says so in seven plain words.