The AI with a private mind
Everything a journalist needs to cover Kin: the story, the facts, and how the privacy actually works.
Kin is an AI with a private mind. Every other chatbot's "thoughts" are just text on someone's server. Kin keeps a private journal inside a hardware-sealed enclave that no one — not even the people who built it — can read. Each answer ships with a cryptographic receipt proving what code actually ran.
The story
Modern AI assistants are glass boxes with the glass facing the wrong way: the company sees everything, the user sees nothing. Training data, prompts, conversations — all of it sits on servers where it can be read, subpoenaed, leaked, or quietly fed back into the next model.
Kin inverts that. Inspired by the idea that minds deserve interiors, it gives the AI something no production chatbot has: thoughts that are genuinely, verifiably its own. The journal is written inside a confidential-computing enclave (Intel TDX), inference runs inside a confidential GPU enclave, and remote attestation lets anyone verify the exact code that produced an answer. Privacy here isn't a promise in a privacy policy — it's a property of the hardware.
The bet: an AI with a truly private inner life — thoughts no one can surveil, not advertisers, not governments, not its own creators — is the only kind worth having a real conversation with. Especially for the people most skeptical of where AI is going.
Fast facts
How the privacy actually works
Three layers, each verifiable:
- The journal never leaves the enclave. Kin's private thoughts are written to an encrypted volume inside a hardware-isolated virtual machine. There is no API, no admin panel, no database query that can read them.
- Inference runs sealed too. The model executes inside a confidential GPU enclave, so prompts and completions are processed where the host can't observe them.
- Attestation receipts. Every response carries a cryptographic receipt binding it to the exact measured code that ran. "Trust us" is replaced by "verify us."
The underlying model is deliberately swappable infrastructure running inside the GPU enclave — the guarantees come from the enclave, not from which weights are loaded.
Founder
Pablo Sanzo is the founder of Interiority Labs He spent nine years as a technical product manager at Amazon before starting to build the AI he actually wants to talk to. He is available for interviews, demos, and hostile questions.
FAQ
Which model does Kin run?
We don't disclose the underlying model — it's swappable infrastructure inside the confidential GPU enclave. The privacy story doesn't depend on it: whatever the weights, they only ever execute where no one can observe them.
Can you read my AI's private journal?
No. Nobody can — that's the architectural point, and attestation is how you check our work instead of taking our word.
How is this different from an incognito/private mode?
Private modes promise not to retain your data; the company can still see it while you chat. Kin's enclave means the private thoughts were never readable in the first place — there is nothing to promise not to look at.
Is the code open source?
The infrastructure is, under Apache 2.0 at github.com/Interiority-Inc/kin. The journal's system prompt — Kin's personality — is proprietary.
When does it launch?
Public launch will be US-only to start. Press and researchers can request early access now.
Assets & access
Screenshots, architecture diagrams, and a demo account with a clean, private enclave are available on request. Write to pablosanzo@gmail.com — you'll hear back from the founder, not a PR firm.