AI proposes. People and policy decide.

1. AI is a governed principalEvery model and agent has an identity, a delegator and explicit limits, and its requests pass the Authority Kernel like anyone else's.
2. Human decision on high-consequence actionsAI cannot hold authority over actions whose consequences are serious or irreversible.
3. Data stays where its classification allowsRequests are routed to models by data classification. Sensitive data is processed only by models inside the customer boundary.
4. Every recommendation is traceableInputs, model, version and output are recorded as evidence alongside the decision they informed.
5. Evaluated before trustedModels are measured against mission tasks before use and monitored after, with results kept on the record.

We design for compatibility with international humanitarian law, the principle of meaningful human control and applicable AI regulation, including the European Union AI Act where it applies.

We decline uses that are inconsistent with these principles.