Adversarial cyber ranges
Realistic, procedurally generated environments with ground-truth attack paths, credentials, and decoys.
CRUCIBLE
Mission-grade evaluation for AI in cyber operations.
CRUCIBLE is an adversarial cyber range that measures how AI systems perform against real attack paths — before they are trusted in the field.
- ATT&CK-aligned
- Autonomous & human-in-the-loop
- Reproducible verdicts
What we do
The proving ground for AI in cyber.
ATT&CK-aligned evaluation
Scenarios and scoring mapped to MITRE ATT&CK, so results speak the language operators already use.
Autonomous & human-in-the-loop
Evaluate AI agents operating on their own, and human/agent teams, as distinct and comparable measurements.
Reproducible verdicts
Seeded, repeatable runs and a clear operator console that turn agent behavior into defensible results.
About
Why CRUCIBLE
As AI moves into cyber operations, the decisive question is not whether a model can act — it is whether it acts within the rules, toward the mission, and without causing harm. CRUCIBLE exists to answer that question rigorously. We give operators and decision-makers the evidence they need: how an AI performs against real adversary tradecraft, where it stays in bounds, and where it does not.