About

Cignal builds AI for the images that matter most and are hardest to read. We started with a question from the Department of Homeland Security: how do you train a detection model on a threat that has never come through a checkpoint? Real screening data captures what actually happened. It says nothing about what hasn't happened yet, and in security, that's the case that matters.

Testing was worse. Evaluating a detection model means staging real threats in real bags. Slow, expensive, dangerous, and it still only covers the configurations someone thought to build. Nobody constructs a thousand variants to find the one that gets through.

So we built a way to make those images. Four phases of DHS's Silicon Valley Innovation Program later, we had a patented voxel-tensor engine that renders physically accurate X-ray, CT, millimeter wave, and multi-spectral scenes from a natural language prompt. Prototype a detector in an afternoon. Test it against cases nobody staged.

Along the way we learned we'd built something bigger than a data generator. People saw images coming out and assumed images were the product. They weren't. What we built was an environment: a place where you describe a scene, see it, and ask what's in it. The images are what falls out of the loop. Agents made that obvious. Today Cignal generates the scene and Ratio reads it back, with reasoning an operator can audit. Both run on the same engine. And if you can generate the case, you can generate the case designed to beat you, which is how testing becomes red-teaming.

Our founders spent their careers in the FBI, in counterterrorism and cyber, looking for what someone worked hard to hide. That's still the job. The images are different.