Governance no longer stops at the front door
In short: until now, governing meant one thing — deciding what an agent is allowed to do inside the tower. By counting who reads us from outside, we understood that the same question applies to any visitor, human or robot: who is allowed in, and how do we prove it.
1. The word “governance” had one meaning
For us, governing an agent means: what it may do is wired in, not written down. A door that refuses, a four-line contract between two agents, a guard that checks the work returned matches what was asked. All of that happened inside, between our own agents.
2. The outside came in through the same door
By counting the robots that visit the site, we noticed the outside poses exactly the same problem as the inside. A robot presenting itself as “GPTBot” and an agent presenting itself as “authorized” are checked the same way: not on their word, on their address, on their proof. Over thirty days, 758 requests out of 2,486 claimed to be an AI robot without coming from the right address. That is not a traffic anecdote. It is an identity check that failed, at the door, like any other.
3. What it changes in practice
Before: governance protected the agents' work among themselves. Now: the same discipline applies to who may read from outside. Three rules, identical on both sides of the wall.
Never take a name at its word. An agent saying “I am authorized” or a robot saying “I am GPTBot”: both are checked against proof, never against their claim.
Measure, don't guess. We don't know how many people read us “roughly”. We know exactly how many real requests from real robots arrived, day by day, page by page.
Refuse cleanly rather than let through out of caution. A visitor who lies about its name is not ignored: it is counted, and turned away.
4. Why this is not just a technical detail
A tower that governs its agents well but lets anyone pass as a known robot has only governed half its house. The front door matters as much as the rooms inside.
Governing a team of agents and governing who may read you is the same rule, applied twice.