Antagon builds the instruments that establish whether an AI system behaves as claimed — scored against a computable optimum rather than a rubric, with the methodology stated and the results reproducible.
Most AI evaluation reports whether a system passed. That answers a different question from the one a programme office actually has, which is whether the reported number supports the conclusion being drawn from it.
Our work is measurement validity: building environments where the correct outcome is computed exactly, scoring behaviour against it, and stating plainly which failures a given metric can and cannot detect.
We do not train foundation models and we do not sell a dashboard. We build instruments, and we publish the methodology so the measurements can be checked by the people relying on them.
A capability statement that hides readiness wastes everyone's time, so each one says where it actually is.
Behavioural evaluation for language-model agents. Places commercial models in instrumented multi-agent environments where the optimal outcome is computed exactly from assigned private values, so each agent is scored against a known optimum rather than a rubric. Two mechanisms; eleven models from eight developers run to date.
A research model that writes a critique of an AI-generated rationale, for a person to review. A held-out evaluation has been run; the model card is the record of its method and results. No accuracy or capability claim is made for it. Open source and self-hosted.
Continuous authorisation and policy-audit automation for IL5/IL6 and FedRAMP High environments. Addresses enterprise data governance, policy enforcement and system-level reciprocity.
Assurance for the agent configuration layer: a versioned schema for an agent's authorisation slice, a multi-model evaluation runner, and a repeatable evidence artifact an ISSO or authorising official can file.
Figures below come from Gauntlet measurement campaigns and regenerate from Antagon's campaign scripts with one command. The scripts are not public; the methodology is available to evaluators on request.
Eleven commercially available models from eight developers, with zero API failures. Every model cleared a conventional 90% market-efficiency threshold — including three that finished worse than a non-participating baseline. An evaluation reporting the environment-level figure alone would have passed all of them. That gap is the measurement problem we exist to close.
| Legal name | Antagon Inc. |
|---|---|
| UEI | KBSGT7CZ4AH3 |
| CAGE | 17E75 |
| Business status | Veteran-owned small business · Non-traditional defense contractor · Wholly US-owned and operated, with no foreign ownership or control |
| Registration | Active in SAM.gov |
| Location | Riverview, Florida, United States |
| Size | 1–10 employees |
| Technology readiness | Gauntlet at TRL 4–5, self-assessed. The basis is the complete system run against eleven commercial models through their production API endpoints, in Antagon's own environment. No independent, accredited or government test, and no deployment, to date. |
| Contact | founders@antagon.ai · (727) 482-8522 |
Everything described on this page was built independently by Antagon, under no government agreement and with no work performed for or directed by a federal customer to date.
Tell us which AI system you are fielding and what decision it informs. We will tell you what Gauntlet can measure about it, and what it cannot. W. Alexander Ousley, founder and principal investigator, answers technical questions directly.
founders@antagon.ai