Verax is an independent range where AI and autonomy systems face live adversaries, unscripted, before they are fielded.
Most assurance happens inside a simulation written by the team that built the system. We test against an adversary who never saw the script.
Red teaming
A team whose only objective is to defeat the system under test. They adapt between runs. A result that holds on the first attempt and collapses on the fourth is the result worth having, and it is the one a scripted harness will never produce.
Benchmarking
Cross-contractor evaluation under identical conditions. Vendors see their own results in full and the comparison in aggregate.
Assurance
What was tested, under which conditions, and what remains unknown.
Scheduled range time against live opposition, with a written finding set you own. You see every run, including the ones that go badly, before anyone else does.
Request a test slotEvidence gathered by a party with no stake in the outcome, structured for NATO TEVV review and comparable across suppliers.
Partner with the range
A Verax mark records that a system was run at the range against opposition it did not control, and that the method and the findings are both on file.
It does not certify that a system is safe. Nothing tested once against one adversary can carry that claim, and a range that offered it would be selling reassurance.
Tell us what the system is meant to do. We will tell you how we would try to stop it.