No. The decision comes from written rules and the authority they grant, not from a model’s judgment. Given identical inputs and identical prior state, the decision and its record are content-identical every time. An AI can propose an action; it can’t approve one.

If something the rules need is missing, the action does not proceed. The system denies the action or sends it to a person for approval. It never fills the gap with a guess.

Monitoring tells you what happened after the action ran. Triarcus decides whether the action may run at all, before it executes. Both are useful, and they answer different questions: monitoring answers “what did the AI do?”, and Triarcus answers “was it allowed to?”

General policy engines are good at evaluating rules you write. Triarcus is built for the authority question: who is allowed to authorize this specific action, under which limits, and what happens when those limits are reached. It also produces a decision record that can be audited afterward. Triarcus can sit alongside existing policy tools rather than replace them.

Not yet. We built two separate implementations from the same written specification, and they agree on every outcome across 10 scenarios, 9 rules and 4 authorities. Both are ours, so the next step is independent testing by an outside party.

Not yet. Triarcus is in the pre-deployment stage, and we’re looking for pilot partners. Contact us →

The core methods are the subject of filed patent applications.