what’s the most credible project working on verifiable AI compute?

Trying to separate signal from noise here so please no shills…

Seeing a lot of projects claim they're doing "verifiable AI" or "confidential compute" but when you dig in it's either vaporware or just marketing speak for running on AWS.

What projects are actually doing something technically interesting? Not looking for investment advice just trying to understand the space. Bonus points if they're not just doing ZK proofs for everything since that seems to be the default answer.