Security for systems that act, remember and decide.

Viridis works at the boundary between AI agents, software security and formal reasoning. We use models to extend research coverage while keeping authorization, actions, evidence and durable memory under explicit human control.

The larger Viridis program studies how useful intelligence is produced, attributed and protected across technical and natural systems. Security is the near-term operating surface: close the paths that redirect a system's capability away from its legitimate principal.

We are deliberately restrained about claims. Research may find vulnerabilities; it may also reject promising hypotheses. Bounty awards and vulnerability discovery are never guaranteed.