Research & Insights

Field notes on adversarial AI

Our thesis, threat research, and product deep-dives, grounded in the attacks that matter for predictive systems.

Testing Models You Can't See Inside: Gradient-Free Robustness for Any ONNX File

In practice a team hands you a model file and a shrug, no task description, no class list, no gradients. Autonomous robustness testing has to start from that reality, which is why it leans on attacks that need only forward passes.

Jun 20, 2026Read

Thesis9 min read

Attack to Defend: Why Predictive AI Needs Its Own Red Team

A decade after adversarial examples were discovered, the models steering sensors, logistics, and autonomy are still shipped without adversarial testing. The gap between what the literature proves and what industry practices is the opportunity.

May 12, 2026Read

Research11 min read

From Pixels to Print: Physical Adversarial Attacks on Overhead Detection

A digital perturbation that works at one exact scale and angle is a curiosity. Expectation over Transformations is what turns it into a patch that survives altitude, rotation, and lighting, and into a real failure mode for overhead perception.

Apr 2, 2026Read

Defense10 min read

The Defender's Playbook: What Actually Makes a Model Harder to Fool

Adversarial training, certified smoothing, and runtime detection each buy something real, and each cost something real. The mistake is treating robustness as a switch rather than a set of engineering trade-offs.

Mar 8, 2026Read