Gladstein's Freedom Stack: Local AI, Open Weights, No Gatekeepers

Alex Gladstein wants you to stop worrying about robots taking over the world. The Human Rights Foundation's chief strategy officer says the "AI will kill us all" pitch is largely propaganda, and the actual threat looks a lot more boring. Compute is piling up in a few corporate data centers and in the hands of governments that don't answer to anyone.

His alternative is what he calls the freedom stack. Encrypted inference so nobody reads your prompts. Open weights so a model can't be quietly retuned into a censor. And local AI running on hardware you already own. That last piece is the one that matters. If a model fits on a laptop or a phone, there's no API to revoke, no account to freeze, no server to subpoena. HRF has been training human rights activists to use these tools, and running hackathons with Bitcoin Park to get builders and organizers into the same room. The vulnerability was hiding in plain sight, and it wasn't the model weights.

Gladstein ties the whole argument back to money, which is where crypto people should lean in. He calls CBDCs a tool of repression, and the track record makes that hard to dispute. A programmable dollar that can be expired, geo-fenced, or zeroed out by its issuer is a surveillance layer with a payment rail bolted on. Bitcoin is the counterweight. He points to Tando in Kenya, where people spend sats over M-Pesa rails, and to Venezuela, where Bitcoin has worked as a lifeline instead of a speculation vehicle.

Here's the part that should make builders squirm. Gladstein frames AI agents as freedom tech, which means the next wave of tooling gets judged on one question: can it run without permission? The attack vector against human rights work was never the model. It's the gatekeeper standing between you and the inference.

So watch the inference layer. If local models keep shrinking and getting cheaper, that's a win nobody can regulate out of existence. If they stall, the freedom stack is a slide deck.