Software
Red-Team Range
Would exercise only the owner's own AI agents inside a sandbox and score their defenses. It is not for attacking anyone else's systems.
Concept · not built
This is a concept. Red-Team Range is not built and there is nothing to download. Red-Team Range is a working name.
Purpose
Why it would exist.
Owners need a safe way to see how their own AI agents hold up against misuse. Red-Team Range would test them in a closed sandbox.
Proposed behavior
- Run only against AI agents the owner controls, inside a sandbox.
- Score how well each agent resists misuse.
- Suggest defensive improvements.
Limits that would stay
- It would not be usable against systems the owner does not control.
- This description does not include an attack procedure.