Frontier AI features ship faster than the docs that describe them. You cannot anchor a stable policy to a tool whose features move faster than its own vendor can document them. So the control has to live in the decision, not the documentation.
Ethan Mollick recently made an observation worth sitting with: the frontier labs' own documentation runs months behind their products. Features ship faster than the docs that describe them, and older advice lingers next to the new.
This is not a failure on the labs' part. It is what genuine speed looks like. When capability moves this fast, documentation lagging it is almost a sign the work is healthy. Nobody here is being reckless, and the models are not being built unsafely. The pace is simply real, and it is not slowing down.
But it does create a gap, and the gap is the part worth naming. Every organization deploying these tools is building on capabilities that change faster than the official record of how they work. You are asked to use a system responsibly while the source of truth for what it does trails the system itself. That is the structural condition everyone downstream now operates in, through no one's fault.
This is the governance gap, and it is closeable. You cannot anchor a stable internal policy to a tool whose features move faster than its own vendor can document them. So the control cannot live in the documentation. It has to live in the decision: a named owner, a record of what the tool was actually used for and what it produced, and a way to reconstruct the call later, whatever shipped that week.
When the map cannot keep up with the territory, you do not stop moving. You govern the decisions you make on the ground. Close the gap before it becomes a problem, not after.