Spent time inside @OpenGradient 's chat product at chat.opengradient.ai today, poking at the identity-separation claim rather than just taking it at face value. $OPG has been moving — volume hit $357.69M on June 15 when Upbit listed it, a 605% spike in a single day. The token narrative was loud. So I wanted to see what the actual product was doing underneath. #OPG

The framing is "your identity is separated from your AI interactions." And architecturally, that holds. OHTTP splits IP from content, TEE strips both before decryption, remote attestation lets you verify the enclave. That part is real.

But here's what sat with me... the Chat routes to GPT, Claude, Gemini, Grok. Frontier providers. Those calls happen inside the TEE's isolated environment — but they still happen. OpenAI processes your words. Anthropic processes your words. The identity is stripped, yes. Your IP is gone. There's no account tied to the query. What's separated is you from your prompt. What isn't separated is your prompt from the underlying model infrastructure you were probably trying to get away from.

I kept turning that over. It's not a flaw exactly — the architecture delivers what it says. Identity linkage breaks. But the mental model most people bring in ("my sensitive query stays inside this privacy layer") doesn't quite match what's actually happening. The prompt still travels. Just anonymously.

Whether anonymous access to the same frontier models is enough... I'm genuinely not sure where that lands.