I assumed AI privacy was a settings problem — toggle a preference, trust the policy, move on.
Then I looked closer at how most AI chat products actually work. The data leaves your device. The identity travels with the query. The promise is legal, not technical.
What @OpenGradient is doing with $OPG feels structurally different. The encryption happens on-device. Your identity is stripped before anything reaches the model. Privacy isn't a setting you configure — it's enforced by cryptography and hardware. The policy doesn't matter much when the architecture already handles it.
That changes the incentive a bit. Most platforms need you to trust them. This one is built so you don't have to.
What I'm not sure about: whether that actually shifts user behavior at scale, or whether most people simply don't think about this until something goes wrong.
They've also added Image Studio — image generation across Gemini, ByteDance, and xAI models, private by default. And Claude Fable 5 is integrated, alongside Nous Hermes in the private chat layer. The stack is expanding faster than I expected.
Worth noting: users spending credits on the platform become eligible for the S2 OPG airdrop. That's a real-use incentive, not just a holding incentive. I'll watch whether that loop actually drives retention or just a spike.
Worth paying attention to if you care about how AI infrastructure gets built — not just what it produces.