Let’s talk about the week the machine got a receipt for everything it did — and why you still can’t get one for what it did to you.
On the twenty-fifth of August, the Linux Foundation announced it was taking over governance of a thing called TRACE. Trust, Runtime Attestation and Compliance Evidence. Built by AMD, Intel, Microsoft and a company called OPAQUE. Here’s what it actually is, stripped of the acronyms.
For the first time, there’s an open standard that gives an AI agent a tamper-proof record of its own behaviour. Hardware-enforced. Cryptographically verifiable. It binds together where the agent ran, what software it used, which policies applied, what data it touched. The record travels with the workload across clouds. Think of it as a black box, bolted to the agent, that nobody on the outside can edit.
And it’s already moving. Nearly a hundred and thirty-five thousand downloads in ten weeks. The big names behind it. A neutral home at the Linux Foundation. The machine just got its audit trail.
Now, I’m not here to tell you this is bad. It’s genuinely the right direction. Independent, portable, hardware-attested proof that a machine did what it said it did — that is exactly the kind of provenance infrastructure this whole conversation has been pointing at. If you’ve been following what I keep saying about proof over suspicion, this is the machine finally agreeing to carry its own paperwork. Good.
But here’s what nobody’s covering. Look at who the receipt is for.
TRACE exists to give enterprises and regulators a way to verify what ran. The company that deployed the agent gets to check it behaved. The cloud provider gets to prove it enforced policy. The vendor — and this is the part they’re proud of — cannot control what counts as proof anymore, because no single company gets to define it. That’s a real advance. I’m not taking it away.
But the person the agent actually acted on? The applicant. The customer. The patient. The one whose credit, job, or coverage just got decided. They have no standing in this system at all.
The agent gets a receipt it can show the company that owns it. You get a decision you can’t see behind. The machine proves it was well-behaved to its operator. It proves nothing to you. And in the one place where proof would actually matter to a human life — the moment a machine decides something consequential about you — you’re still standing outside the door with nothing.
That’s the asymmetry nobody’s saying out loud. We are building verifiable machines before we built verifiable accounts of what they do to people. The machine got its black box first. The human still has no black box, no receipt, no independent record that travels with the decision that changed their situation.
Here’s what I mean concretely. Say an agent decides your loan application, or an agent moves your insurance claim down the queue, or an agent scores your job application before a human ever looks at it. Under TRACE, the company can hand a regulator a cryptographic receipt showing the agent followed policy, ran in a trusted environment, touched your data the way it was supposed to. Perfectly verified. Immaculate.
And you? You get a notice that a machine was involved. Maybe an explanation. Almost certainly not the receipt. Not the record of what it did with the version of you it was working from. The verification happened. It just wasn’t for you.
The deep problem is that we keep building proof infrastructure for the machine and calling it trust infrastructure for everyone. It isn’t. A receipt that exists to clear the operator is not the same as a record that exists to protect the person. They sound the same. They point in different directions.
So here’s the operator lesson, and it’s a sharp one. The next fight in this whole human-sovereignty story is not whether machines get audit trails. They’re going to. The fight is who gets to demand the trail. When a machine acts on you, the ability to see what it did, how it classified you, what data it used — that right has to sit with you, not just with the company that deployed it and the regulator it answers to. Portable proof that the machine carries for its owner is the first half. Portable proof that travels with you and your case is the second half. And we have not built the second half.
Don’t misread this as anti-machine. TRACE is progress. The problem is it’s progress that makes the machine legible to the powerful and still opaque to the people it touches. It closes the audit gap for the company. It does nothing for the credibility gap on your side of the table.
The human-sovereignty version of this — and this is the part I’d actually fight for — is the right to the receipt. Not the summary. Not the friendly letter. The record. The independent, tamper-evident account of what the machine did, that you can hold and challenge, the way the company can hold its own copy and wave it at a regulator.
Because right now, the machine has a black box it can show to the people who own it. And you’re sitting there, on the receiving end of its work, with nothing to check, nothing to challenge, nothing that even tells you it happened in the way they say it did.
The receipt shouldn’t be a privilege of the operator. It should be a right of the person.
The machine just got proof it behaved. It’s time somebody built the version that proves, to you, what it actually did. That’s not an anti-AI position. That’s the whole point of all of this — keeping the human sovereign inside a system that’s getting very good at making itself legible to everyone except the person it’s acting on.