Cignal Demonstrates Ratio, the First Compositional VLM for Security Screening

REEDSVILLE, PA, July 2, 2026 - Cignal Demonstrates Ratio, the Security Screening Industry's First Compositional Vision Language Model (VLM).

Ratio is a first-of-its-kind AI that analyzes security checkpoint images the way TSA screeners do.   In the below video, you'll see Ratio analyzing a CT image to identify items in a security bin.

‍

Ratio is what is called a compositional Vision Language Model (VLM), or a visual reasoning model.  This means that unlike current AI and deep learning methods that essentially "look once" to determine if a bag contains a dangerous or a prohibited item, Ratio can zoom in, look around, pick apart, and analyze the contents of a bag to identify anomalies or dangerous items. One of the challenges with conventional deep learning methods is brittleness: traditional models trained on images from a specific type of CT machine tend to only work with that OEM's systems.  They also need to be extensively trained to identify variations in threat item shape, texture, or state (disassembled threats).  Consequently, if a threat item is out of distribution (OOD) in terms of the model's training data, we can get a false alarm.

Compositional VLMs do not have these limitations.This means it's going to get a lot harder for terrorists, drug traffickers, and other criminal organizations to ship, carry, or conceal dangerous and illegal items in airports and across borders, because Ratio can use visual reasoning to investigate what it sees.

This represents a significant leap in terms of aviation security capability, and Cignal is excited to help bring these emerging AI capabilities to our customers and partners.  Special thanks to Justin Tornetta and Daniel Mallia for their work on this project and our SPIE ADIX paper ("Let's be reasonable: why compositional VLM's are the future of ATR") that outlined Cignal's VLM concept just a few short months ago.

‍Read Cignal's VLM research paper presented at SPIE ADIX X here.

‍