Introducing TortSignal: How we detected Dexcom’s Hidden Mass Injuries

At WhyHow.AI, we are reimagining what an AI-native plaintiff referral law firm would look like if we rebuilt it from the ground-up with data scientists and engineers. We build dedicated internal tools to help identify new mass tort cases, and acquire and qualify plaintiffs.

On July 10 2025, while building our internal AI intelligence platform, TortSignal, to detect mass tort signals, we detected an overlooked medical device potentially hurting tens of thousands of people. The issue lay with a diabetes sensor, Dexcom G6 / G7 sensor units that were allegedly causing potentially tens of thousands of people to get into serious injuries due to sensor errors failing to notify them of dropping glucose levels.

This sensor manufacturing issue was worth potentially hundreds of millions in settlements, and was at the time, missed by other law firms that were separately pursuing issues related to its receiver.

Combining our experience working in the legal industry and the quantitative trading industry, we use TortSignal to analyze public data and understand where the next large mass tort cases are going to come from.

The Dexcom Insight

While firms have focused on Dexcom's Class I receiver recall (700,000 units with faulty speakers), TortSignal identified a more serious sensor manufacturing violation affecting tens of million sensors (the unit that users attach to their body) from mid-2023 to Q4 2024.

In plaintiff injury and medical device liability - it is not just important that harm is caused, but that you are also able to navigate the legal liability regime to identify an appropriate mass tort signal - pre-emption, 510k regimes, clear evidence of causation, scope of liability, ability to pay, and hundreds of other factors come to mind that can complicate a claim.

Our AI system is built on top of a robust framework of deterministic legal reasoning that is meant to sift through the noise like a domain expert, and surface what is truly relevant.

Combining Multiple Signals in the TortSignal Platform

TortSignal continuously monitors:

• Regulatory actions: FDA warning letters, inspection reports, enforcement patterns

• Adverse events: MAUDE analysis for injury patterns beyond publicized recalls

• Social Media: Patient forum monitoring for quality degradation signals

• Chronology Builder: Timeline analysis linking complaints and regulatory actions

• Existing Litigation: Any existing related litigation and their progress

• FOIA Requests: FOIA requests as part of a way to access proprietary and publicly available data flows

The Dexcom case demonstrates how these capabilities identify opportunities that strengthen overall litigation strategies.

Case Study: Dexcom Manufacturing Deviation Discovery

TortSignal's regulatory analysis combined MAUDE data with FDA warning letters, specifically flagging a manufacturing deviation raised in a March 2025 FDA warning letter revealing unauthorized changes to Dexcom's sensor enzyme coating. The FDA found that Dexcom:

• Failed to file required 510(k) notification for a "significant design change" that replaced a critical sensor component

• Produced sensors that were "less accurate than those with the original component"

• Created sensors that "cause higher risks for users who rely on [them] to dose insulin"

Timeline:

• Mid-2023: Unauthorized coating change begins

• June 2024: FDA inspection flags violations at Mesa facility

• October 2024: Second inspection at San Diego confirms issues

• March 2025: Warning letter published after inadequate responses

This ~18-month manufacturing deviation affecting tens of millions of sensors created more impactful regulatory grounds for claims for the sensors, beyond the receiver recall.

Data Patterns Across Hundreds of Thousands of Claims

MAUDE Analysis: With 500,000+ adverse event reports, including 3500+ reports of fainting, and at least 13 deaths. Majority involve failed glucose alerts - extending beyond receiver hardware issues.

User Reports: Social media analysis showed sensor failure rates of 25-70% in user communities, with specific complaints about:

• Sensors dying before 10-day lifespan

• Inaccurate readings requiring finger sticks despite "no fingerstick" marketing

• Missed alerts during dangerous glucose events

• Quality differences between manufacturing locations (Malaysia vs. US plants)

Corporate Awareness: Users report Dexcom support asking for sensor "country of origin" when handling failures, suggesting internal tracking of location-specific quality issues.

Sample of MAUDE Complaints

Maude complaints: 500,000 complaints, ~3,504 injuries through fainting

How TortSignal is more comprehensive than searching the MAUDE database directly

As compared to what is available for MAUDE, which is a simple keyword search, we are able to pull out 84% more death-related adverse events by understanding clinical terminology variations (finding 'patient expired,' 'fatal outcome,' and 'did not survive' when lawyers search for 'death'). 

Through the power of LLMs and natural language search, we are able to connect medical reports across keywords that are similar but would otherwise be missed in a simple keyword search. For example, across hypoglycemia reports, we want the LLM to automatically aggregate keyword search fragments across terms like 'fainting,' 'syncope,' and 'loss of consciousness’, which TortSignal is able to do at scale.

Of course, most critically, TortSignal identifies hidden patterns like the manufacturing deviation by correlating temporal clusters with FDA warning letters - discovering the $1 billion opportunity others missed.

Example: Semantic search connected MAUDE reports 3004753838-2025-048342 (teen hypoglycemic seizure) with 3019004087-2024-00008 (loss of consciousness) and 3004753838-2024-193928 (rollover crash) - all sharing the pattern of CGM reading falsely high before glucose crisis.

Other Flags - Differences in Consumer Expectation from Advertising

One thing that stood out in our large scale data analysis of social media data was that users reported a difference in the expectation of reliance on the Dexcom devices.

These comments stood in contrast to the idea that Dexcom users could rely heavily on Dexcom devices, despite warning labels that noted limitations of liability and that warned users to do manual calibrations in case they felt symptoms did not match the readings.

Some of the language in Dexcom material highlighted the diminished role of fingersticks (needles that are used to prick the skin to measure blood glucose levels periodically each time a measurement is to be taken).

G7 Manual

G6 Manual

Dexcom Website

Partner vendor advertisements

Mass Tort Case Value Analysis

Given the range of possible injuries that were reported by Dexcom users, a settlement matrix of different injuries was created internally to help track seriousness of claims.

Affected Class: G6/G7 purchasers from mid-2023 to Q4 2024 who bought sensors with unauthorized modifications.

Short Selling on Medical Device Data

After a short-seller released its analysis of the Dexcom data a few days ago in September 2025, Dexcom’s share price dropped by ~10%. Analyzing raw medical device data has traditionally been used by hedge funds to help front-run information flow and to get an edge in trading. 

Tools like TortSignal that read raw MAUDE data and allow organizations to get an understanding of harmful medical devices before others can provide significant commercial leverage.

Multi-Source Intelligence

TortSignal’s strength is in the ability to flexibly combine a range of different data sources to flag signals and connections that would otherwise go missed. Other discovered medical devices typically involve different combinations of data sources. For this Dexcom issue, this specific discovery required cross-referencing:

• FDA warning letters with recall notices

• MAUDE complaint patterns with manufacturing timelines

• Social media quality reports with regulatory violations

Traditional research approaches these sources separately. TortSignal's AI-integrated approach helps combine data from different sources and build stronger connections and correlations which may otherwise get missed.

Collaboration

TortSignal is currently an internal tool developed in collaboration with a number of design partners.

Intelligence Sharing: The use of TortSignal for MAUDE and other data sources to identify and support mass tort cases

Plaintiff Acquisition: Plaintiff acquisition and intake based on cases identified

The Goal: Supporting stronger litigation that maximizes consumer recovery while holding manufacturers accountable for systemic quality failures.

At WhyHow.AI, we are reimagining what an AI-native plaintiff referral law firm would look like if we rebuilt it from the ground-up with data scientists and engineers. We build dedicated internal tools to help identify new mass tort cases, acquire and qualify plaintiffs, and provide real-time data pipelines and case inventory analytics between law firms, litigation funders and marketing vendors.

Connect with us here!