Cybersecurity/United Kingdom; Singapore
They faked the glucose reading and the pump delivered more insulin
A hardware-in-the-loop platform fed false glucose readings to a real insulin controller, and the pump delivered more insulin.
A modular hardware-in-the-loop platform integrating continuous glucose monitors, fluidic systems, insulin controllers, wireless communication and a real-time virtual subject simulator was built to evaluate automated insulin delivery security. A false data injection attack on the glucose sensor led to an increase in insulin dosage delivered by the control algorithm relative to baseline, which if undetected could result in a life-threatening condition.
What happened
An automated insulin delivery system is a closed loop. A continuous glucose monitor measures blood sugar, an algorithm decides what to do, a pump delivers insulin. For people with type 1 diabetes it is a substantial improvement over manual dosing, and it depends entirely on the controller believing the sensor.
Medical device security research has repeatedly shown that the wireless links in these systems can be intercepted and spoofed. It has rarely shown what happens next, because testing that on a person is not possible and testing it in pure simulation does not reproduce the real communication and timing behaviour.
A team at Imperial Global Singapore and the Department of Bioengineering at Imperial College London, with the School of Electrical and Electronic Engineering at Nanyang Technological University, built a platform that closes the gap.
They injected false glucose data. The pump delivered more insulin.
The test
The platform is hardware-in-the-loop: real devices connected to a simulated patient. It integrates continuous glucose monitors, fluidic systems, insulin controllers and wireless communication pathways with virtual subjects.
Physically it comprises programmable peristaltic pumps, 3D-printed flow cells, glucose sensors and a real-time virtual subject simulator. The fluidics reproduce physiological glucose dynamics so that the sensor is measuring something real, which is what preserves the authentic sensor behaviour and timing.
The authors identify the gap this fills precisely: there are limited assessment tools that preserve the communication and timing paths of continuous glucose monitor sensors within an automated insulin delivery system. Pure software simulation skips those paths, and they are where the attack lives.
The platform was validated by demonstrating reproducible glucose trajectories and closed-loop operation with an open-source insulin controller, before the attack scenario was run.
The numbers
| Element | Detail |
|---|---|
| Platform type | modular hardware-in-the-loop |
| Components | programmable peristaltic pumps, 3D-printed flow cells, glucose sensors, real-time virtual subject simulator |
| Controller | open-source insulin controller, closed loop |
| Validation | reproducible glucose trajectories |
| Attack tested | false data injection on the glucose sensor |
| Effect | increased insulin dosage delivered relative to baseline |
| Consequence if undetected | authors state could be life-threatening |
The paper reports the direction of the effect rather than its magnitude in the abstract, which is worth stating plainly: we can tell you insulin delivery increased, and we cannot tell you by how much.
The result
The platform generates reproducible glucose trajectories and supports closed-loop operation through integration with an open-source insulin controller – the validation that the apparatus works before any attack is attempted.
A false data injection attack scenario on the glucose sensor led to an increase in insulin dosage delivered by the control algorithm relative to baseline. The authors state that if undetected, this could result in a life-threatening condition.
Their impact statement is direct: this provides evidence that automated insulin delivery systems vulnerable to cyberattacks pose substantial risk to patient safety, underscoring the need for improved regulation and design of personal wearable medical devices through cyberphysiological system modelling.
How this compares to what came before
Insulin pump security has a long history, including well-publicised demonstrations at security conferences more than a decade ago showing that pump commands could be spoofed. Regulators responded, manufacturers improved authentication, and the issue was treated as addressed.
Those demonstrations targeted the pump’s command interface. This work targets the sensor instead, which is a different and arguably harder problem: the controller must trust its sensor to function at all, so there is no obvious equivalent of refusing unauthenticated commands.
The methodological advance is carrying the attack through the full loop. Showing a protocol can be spoofed is one claim; showing the spoof changes the dose delivered by the pump is the claim that establishes clinical risk.
Separately, a systematic review in this same batch catalogued 316 patient-level harms from hospital cyberattacks and could produce no incidence rate because reporting does not exist. This paper is the complementary piece: a controlled demonstration of a mechanism, where the field has mostly had after-the-fact narratives.
What it means
The structural problem the paper exposes is sensor trust. A closed-loop controller has no independent way to know whether its glucose reading is real. Authentication on the wireless link helps; plausibility checking on the data helps; neither is a complete answer, because a sufficiently careful attacker can produce readings that are authenticated and plausible.
For regulators, the framing matters. Device security assessment has focused on the device. What this platform tests is the cyberphysiological system – device, communication, control algorithm and physiology together – and the authors argue regulation should follow.
For people wearing these systems, the appropriate response is not alarm. This is a laboratory demonstration against a research controller, not a report of attacks in the wild, and the clinical benefit of automated insulin delivery is large and well established.
For manufacturers, the useful output is the platform rather than the attack. An apparatus that can safely test an attack through to the delivered dose is what the field has lacked.
Business ideas from this paper
Security testing platforms for closed-loop medical devices. The hardware-in-the-loop approach generalises to any device that senses and acts on a patient: artificial pancreas systems, closed-loop anaesthesia, cardiac devices. Building and selling those test rigs is a defined product.
Sensor plausibility monitoring. A layer that checks glucose readings against expected physiological dynamics and flags implausible trajectories is a software addition that addresses the specific attack demonstrated here.
Regulatory test standards. If assessment should cover the cyberphysiological system rather than the device, somebody has to write the standard and the test protocol. That is pre-competitive work with regulatory demand.
Independent device security assessment for medtech. Manufacturers currently self-assess. An independent laboratory with a validated hardware-in-the-loop platform would serve both regulators and insurers.
What could be wrong with it
The subject is virtual. Glucose dynamics are reproduced by pumps and flow cells against a simulation, so the physiological model carries its own assumptions and no person was dosed.
The magnitude of the insulin increase is not given in the abstract, which makes the severity hard to judge. Direction is established; size is not.
One controller, one sensor type, one attack scenario. This demonstrates a class of risk, not a survey of deployed commercial systems, and the systems people actually wear were not tested.
The open-source insulin controller used may have different protections from commercial devices, in either direction.
No independent replication.
How sure can you be?
We score this 7 out of 10.
It is a peer-reviewed paper in BME Frontiers, using a built and validated hardware-in-the-loop platform with real pumps, sensors and fluidics, with named institutions at Imperial College London, Imperial Global Singapore and Nanyang Technological University.
It loses points because the evidence is a laboratory demonstration rather than a controlled study or a survey of deployed devices, there is no independent replication, and the data are not publicly deposited.
Carrying the attack through to the delivered dose is what makes this more than a protocol analysis. The missing magnitude is the detail we would most want before drawing conclusions about severity.
Do this today
If you design closed-loop medical devices, the question to ask is what your controller does when the sensor lies. Authentication answers a different question, and the controller still has to decide whether to believe a reading.
If you regulate medical devices, the cyberphysiological framing is the contribution. Assessing the device in isolation misses attacks that only matter when they reach the patient.
If you use an automated insulin delivery system, this is laboratory research and not a report of attacks in the wild. Keep your device software updated, which is the standard advice and remains the right advice.
Source: Toward Trustworthy Diabetes Technologies: A Dynamic Platform for Safety and Security Evaluation, BME Frontiers, September 2026. PMID 42819556 · doi.org ().
Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.
- The platform preserves the communication and timing paths of real continuous glucose monitors within a closed-loop system.
- Built from programmable peristaltic pumps, 3D-printed flow cells, glucose sensors and a real-time virtual subject simulator.
- False data injection on the sensor increased delivered insulin relative to baseline.
- The subject is virtual, so no person was dosed - and the physiological model carries its own assumptions.
A modular hardware-in-the-loop platform was built and validated with real pumps, flow cells and glucose sensors in closed loop with an open-source insulin controller.
Questions people ask
What is automated insulin delivery?
A closed-loop system where a continuous glucose monitor measures blood sugar, an algorithm decides how much insulin is needed, and a pump delivers it. It removes most manual dosing decisions and is life-changing for many people with type 1 diabetes.
What is false data injection?
Feeding the controller fabricated sensor readings. If the system is told glucose is high when it is not, it will respond by delivering insulin that is not needed.
Why is extra insulin dangerous?
Too much insulin causes hypoglycaemia, which can progress to seizure, unconsciousness and death. The authors note that undetected, the attack could result in a life-threatening condition.
What is hardware-in-the-loop?
Real hardware - pumps, sensors, controllers - connected to a simulated physiology rather than a person. It preserves the real communication and timing behaviour while making the experiment safe.