Research

What we publish

Cybersecurity for AI

Responsible disclosure

We've identified confirmed vulnerabilities in widely-used open-source AI/ML projects, including critical-severity findings, and coordinate with maintainers before public disclosure.

AI for Cyber

Edge-deployable autonomous reasoning

Patent pending

Compact neural architecture for autonomous reasoning on CPU-only edge hardware, with no network or external dependencies.

AI for Cyber

Vectorless RAG

Retrieval system using hierarchical topic indexing instead of vector embeddings. Works without embedding models or vector database infrastructure.

AI for Cyber

Tiny recursive models

Recursive neural models small enough to run inline on a device or collector, deciding on each event in real time. Adaptable to any domain, they cut noise and reduce data at the source, dropping what doesn't matter before you pay to move and store it.