Subscribe now

Three stories this week look unrelated and are the same story. A piece of evidence that may have been made by a machine. A frontier model whose safety lived inside the model, then had its access pulled by a government. And the loudest voice in the industry calling agents the next platform. Read through one lens, each points at the same missing layer: proof.

1. Evidence.
   A police officer is under investigation for allegedly using AI to create evidence, reportedly the first case of its kind in the UK. Take no view on the case, the structure is the lesson. Once an artifact can be generated on demand, its existence proves nothing about who made it, or when. Detection guesses after the fact, and guesses worse each model cycle. The answer is provenance carried by the artifact itself, a record made at the moment of creation of what produced it, a human or a machine, that anyone can check without trusting the holder. The first is an argument. The second is evidence.

2. Containment.
   A frontier model shipped this week with its safety built into the model itself, and then had its access restricted by a government order. Two different forces, one architectural lesson. When safety lives inside the model, both the model and its safety can be talked out of, switched off, or revoked by whoever controls access. The layer that holds sits outside the model. It holds no handle the model can reach. It fails toward stop, not toward open. And it keeps a signed record of every decision, so the safety itself can be checked.

3. Agents.
   The agents are coming, and they will not stay in the chat window. They will take actions at machine speed, payments, code, records, handoffs. Every action is a new thing that can go wrong, and most agents ship with no external way to stop one mid-step. The layer that completes the platform is a brake on the action path, outside the agent, that defaults to stop and records what it did. Build the agents. Also build the thing that can stop them.

The through-line.
Across all three, the missing layer is the same. Proof of what acted, made at the moment it acted, that anyone can verify without trusting the source. That is what we build at Regulayer, proof infrastructure for AI. Regulayer for what an AI did. HumanMark for what a human made. Together, the Signed Internet.

Don’t take my word for any of it. Verify a real cryptographic receipt yourself at regulayer.com/verify. It takes about 300 milliseconds, in your browser, with no contact with us.

Proof, not promises.