Answers are grounded in this paper and cite their sources.
More than half of the traffic on the internet now comes from machines, and the share grows every year. The gates that decide which machines get through can tell which company runs an agent, and that is all they can tell. They can't say whether the agent is cleared to fill out a form, move money, or act without a person watching, and they can't see how it has behaved before. So every website makes its own trust decision alone, with almost no information, and the agents that want to wreck things get the same answer as the agents that want to help. This paper argues that no AI agent should act online without a certificate being checked, and that the check should happen at the places every agent already passes through: the model APIs that give agents their brains, the content delivery networks in front of most websites, the browsers and operating systems that agents act through, the protocol that signs their requests, and the payment rails. A certificate, in the sense that Licensed to Act lays out, says what an agent is cleared to do and how it has behaved. An agent without one can read the web, but it can't act on it. An agent with one acts within its scope, and an agent whose record stays clean earns room to act on its own. The whole method rests on one principle: misalignment has to cost more than alignment, everywhere and at every moment, and the gates are where that price gets charged. Gating the whole internet can happen without anyone being in charge, because it has happened twice before, when email adopted sender authentication and the web went encrypted. The paper also faces what gating can't do: it won't stop a well-funded attacker on its own, it depends on telling people from agents, and it hands real power to whoever runs the gates. It argues the answer is an open standard, many certifiers, and a floor that nobody falls below. Humanity Systems working paper. Part of a series on one question: how do we trust AI, now and into the future?
No takes yet. Share an insight, caveat, or question.
A TUCKER (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: