What happened

Abbott’s AI-supported continuous glucose monitoring (CGM) sensors, used by people with diabetes to automatically track blood glucose levels, were linked to seven reported deaths after devices allegedly delivered incorrect or delayed readings. 

These failures appear to have occurred across multiple locations where the sensors were in use, leading to undiagnosed hypoglycaemia or hyperglycaemia, missed alerts, or inaccurate data that led users to administer the wrong insulin dosage. 

The incident suggests potential direct harms, including life-threatening glucose mismanagement, and indirect harms, such as loss of trust in essential medical technologies and increased anxiety among patients dependent on automated monitoring. 

Why it happened

The core issue was a manufacturing defect traced back to a single production line responsible for producing the affected sensors. 

• Root cause: Abbott stated the cause was a manufacturing defect that has been identified and resolved. This indicates a quality control failure in the production process of the physical sensor component.

• AI system malfunction: The sensor is part of an AI system, which processes the sensor data to infer and report glucose levels and trends. The manufacturing defect caused the input data to the AI system to be flawed, resulting in an AI system malfunction (inaccurate output) that directly caused or contributed to the adverse health outcomes.

• Transparency and accountability:

  • The incident highlights the critical need for robust post-market surveillance for medical devices, particularly those incorporating AI, as a manufacturing flaw can have lethal consequences.

  • While Abbott initiated a "medical device correction" and offered replacements, the fact that 7 deaths and hundreds of severe adverse events were reported globally before the company publicly announced the issue and its correction raises questions about the timeliness of detection, reporting, and corrective action across all distributed regions.

  • The reliance on AI to interpret sensor data means that even small physical errors can be amplified into life-threatening misdiagnoses by the system's output, emphasising the need for transparent safety mechanisms and fail-safes to prompt users to check readings against symptoms or a traditional blood glucose meter when readings are anomalous.