Concept: closed loops that act on ketones

This entry used to argue for a sensor. That argument has been won: Abbott's Libre Duo reads glucose and ketones from one wearable, every minute, took a CE Mark in Europe in May 2026, and the 10-day version received FDA De Novo authorization on 25 August 2026 — authorization, not a launch, and still not a loop that acts.²³⁵ So the concept moves up a layer, to the half that nobody has built.

The remaining gap: nothing is listening. When an infusion site fails or illness strikes, ketones (beta-hydroxybutyrate) can climb toward diabetic ketoacidosis — sometimes while glucose still looks reassuring. In a reported case, a wearable ketone monitor caught exactly that: sustained ketosis above 1.0 mmol/L for 14 hours while CGM showed normal glucose.⁷ Today the best a dual sensor can do is tell you. As of this cutoff, no commercial automated insulin delivery (AID) system reads a ketone channel, and no published standard defines what a loop should do when it sees one rising — increase basal, prompt an injection and a site change, refuse to trust the pump, or simply escalate loudly.⁴ A warning light is only worth as much as the thing wired to it.

What we want built. A loop that treats ketones as a first-class input: it notices ketones rising while glucose creeps up, infers "this insulin is not getting in", stops quietly compensating, and tells you to change the site and inject by pen — the reasoning a person does today, done automatically and potentially hours earlier. The prerequisites are a published accuracy figure for the ketone channel (you cannot let an algorithm act on a signal whose error bars are unpublished), an agreed action policy, and a regulator willing to clear it. As of this cutoff, no published accuracy figure, agreed action policy, or clearance for such a loop exists.

Who is actually building the sensing side. Abbott led first with the Libre Duo — CE-marked, then US-authorized as a 10-day sensor, still not a commercial launch as of this cutoff.²³⁵ PercuSense was reported as starting two Helmsley-funded type 1 diabetes studies in summer 2026, measuring glucose, lactate, ketones and tissue oxygen from a single probe.⁴ Biolinq is reported to be adding lactate, with ketones and cortisol under consideration.⁴ Dexcom has said it will ship potassium first (around 2027-2028) and ketones after.⁴ Sequel has a public agreement to connect its twiist AID system to Abbott's dual sensor, and Abbott has announced plans for twiist and iLet pairings by the end of 2026 — the closest anyone has come to closing this loop, and still a company plan rather than a product, with no published rule for what the loop would actually do about a rising ketone.⁶⁵ The multi-analyte hardware race is real and crowded; the algorithm race has not started.

The cited review reports an international expert consensus calling for precisely this class of sensor, noting poor adherence to fingerstick ketone testing and the risk of ketosis at normal glucose levels with SGLT2 inhibitors.¹ A registered McGill trial (NCT06753994) is studying continuous ketone monitoring in people with type 1 diabetes using SGLT2 inhibitors.⁸ We keep this as a concept to hold the unbuilt half visible — the sensing is arriving, the intelligence is not.