Made in the USA

Red Teaming for Vision AI.

Identify exploitable vulnerabilities in how machines see

Governent validated

Validated through a 4-phase DHS Other Transaction Agreement

Vision as vulnerability

Every autonomous system sees through a model nobody has attacked.

For Intelligent Machines

Deployed Abroad. Defending at Home.

Autonomous systems in the field. Fixed screening infrastructure at home. The security of the vision and sensing systems in these machines are critical to our national security and defense.

Machine vision is mid-adoption. There is no red team tooling. Testing physical machines is slow, expensive, sometimes dangerous, and it only ever covers what someone thought to bring to the test. An adversary doesn't work that way.

That's why we built Cignal Engine through the U.S. Department of Homeland Security Silicon Valley Innovation Program (SVIP).

Read the DHS Press Release about Cignal at SVIP demo day

AI Defense

See It. Train It. Break It. Fix It.

Cignal Engine generates the attack. It's a patented voxel-tensor environment that generates high-fidelity non-natural imagery.
‍
That's what makes the attacks real. Physics-accurate generation, not perturbed pixels that fall apart the moment they meet hardware.

Cignal breaks the model, explains why it broke, so vision AI is hardened against offensive visual exploits.

Built and validated through four phases of DHS SVIP.

Learn more about Cignal AI

Emulate adversaries

Model the adversary, not just data.

Identify Vulnerabilities

Know why it broke, not just that it did.

Offensive AI

Exploiting Visual Systems

Exploit gaps in training data to identify new camoflauge against UAS, evaluate different data poisoning strategies, and reverse-engineer the sensors and targets an adversary's models were trained to find.

Automated Annotation

Generate fully-labeled image streams.

Adversarial Emulation

Model adversarial intent, not just data.

Multi-Modal GenAI

Language & multi-spectral data generation.