Where it matters that the answer is real and the operator is blind, the network's guarantees become the product.
Run sensitive workloads on the Confidential tier. The operator is blind inside the enclave and every reply carries a hardware-attested receipt, so legal and security teams get proof, not promises.
Keep inference on attested hardware in a chosen jurisdiction, with a per-reply quote that binds the running code and model. Auditable by design, without trusting an operator's word.
Agents act on model output, so the output has to be real. OGONG checks each step came from the model you asked for, which closes the door on a swapped or degraded model corrupting a long chain of actions.
Serve any open model on the Verified tier. Correctness is proven by re-check, with no gatekeeper deciding which models are allowed, and no provider able to quietly substitute a cheaper one.