Most AI platforms treat verification like a black box. The system picks a method, you receive an output, and you never see what actually stood behind it.
OpenGradient's trust menu operates differently. No single method imposed. Users choose between TEE, ZKML, and Vanilla signature verification depending on what the inference actually requires.
The closer comparison: choosing your own auditor instead of receiving someone else's bill of health by default.
Here is the trade-off OpenGradient accepted. A single hidden verification method gives users one outcome to receive — trust the platform's choice or don't use it. The user never needs to understand what's backing the result.
The trust menu replaces that single outcome with a decision. Your guarantee is simultaneously hardware-rooted speed if you pick TEE, cryptographic certainty without trusting hardware if you pick ZKML, and minimal overhead with a thinner guarantee if you pick Vanilla — a choice you actually have to understand to make correctly.
That choice is yours to carry whether you engage with it or not.
If you pick the method that matches the inference's actual stakes, you get verification that fits the risk. If you default to whichever is fastest without understanding what you gave up, the gap you ignored going in is the same gap deciding what happens when that result turns out wrong.
A trust menu is a more structurally honest model than a platform that silently picks one method and calls it security.
OpenGradient built a system for people who want to choose what they trust and why. Whether people who want that choice and people who just want a fast result are the same user is what the data will show.