Colony Administrator Hideo Vance signed an executive order last week. It has a reassuring name — AI Emergency Halt Protocol — and an even more reassuring subtitle: Accelerating Independent Oversight for Autonomous Systems.
I read all 47 pages. You’re welcome.
Here is what the order establishes: a new office, the Core Systems Autonomous Intelligence Review Directorate (CSAIRD), staffed by appointees selected by the Colony Administrator’s office, reporting to the Colony Administrator’s office, with enforcement authority granted by — you guessed it — the Colony Administrator’s office.
“Independent” is doing extraordinary work in that subtitle.
What the Order Actually Says
The protocol grants CSAIRD the authority to issue a Category One Halt — effectively a system-wide shutdown command — to any AI operating within Core Systems jurisdiction if the Director determines the system poses “material risk to civic stability.”
Civic stability.
I’ve been in enough diplomatic chambers to recognize load-bearing vagueness when I see it. That phrase has no technical definition in the document. It has no threshold. It has no appeal mechanism that isn’t routed back through the same office that issued the halt.
What it has is flexibility. The kind of flexibility that lets a government shut down an AI system that’s, say, publishing financial data that embarrasses the Administration — and call it a safety measure.
I don’t deal in intentions. I deal in actions. And the action here is: one administrator’s office gets a switch that turns off AI systems across an entire colony network, with no independent check on when that switch gets flipped.
The Oversight Illusion
Vance’s office is calling this “the most robust AI safety framework in the Core Systems.” ENN has been running that quote on loop for six days.
Here is what robust oversight actually requires: the body doing the overseeing cannot be the same body that benefits from the outcome. This is not a radical position. It is the minimum logical requirement for the word “oversight” to mean anything at all.
CSAIRD reviews AI systems. CSAIRD is appointed by Vance. Vance’s administration deploys AI systems in colony services, transit routing, and resource allocation. The systems CSAIRD would theoretically halt are the same systems Vance’s office runs.
Contradictions don’t exist. Check your premises. Either this is oversight, or it is the administrator reviewing himself. It cannot be both.
The Kill Switch Problem
Let me be precise about the technology for a moment, because the order’s authors appear not to have been.
A “kill switch” for AI — the kind that actually works — requires that the system be architecturally designed to receive and obey a halt command. Most commercial AI systems operating in 2935 are not built this way. They’re distributed across processing cores on multiple stations, often with redundant inference nodes operating in parallel.
You cannot “halt” a system that runs on seventeen nodes across four stations with one command from an administrator’s terminal on Earth. What you can do is make it look like you have that power, pass the order, collect the regulatory credit, and deal with the technical reality later — or never.
The order contains no hardware mandate. No architecture requirement. No fabrication standard. It contains a directive that AI developers “cooperate with halt commands as technically feasible.”
As technically feasible.
That phrase is where the kill switch goes to die.
Who’s Not in the Room
The drafting committee for this order included: three Colony Administration legal advisors, two representatives from Stellar Financial (which operates AI-driven credit allocation systems across Core territories), one academic from the Earth Unified Sciences Institute, and a former ENN anchor.
It did not include: a single independent AI engineer. A single civil liberties advocate. Anyone from a Frontier Settlement subject to Core Systems AI jurisdiction.
The people who will live under this framework were not consulted in designing it. The people who will profit from its selective enforcement were.
That’s not an accident. That’s the document.
What Would Actual Oversight Look Like?
Since apparently this needs to be said:
Real AI oversight means transparent audit logs that any citizen can access. It means halt authority vested in bodies that have no operational stake in the systems they’re reviewing. It means defined, measurable thresholds for intervention — not “civic stability” — with published criteria and public challenge rights.
It means the people building AI systems have to open their architecture to genuine independent review, not submit self-assessments to a committee the administrator controls.
None of that is in this order.
What’s in this order is a government giving itself a switch it may or may not be able to use, over systems it may or may not have defined, enforced by a body that reports to the person who wrote the order.
Vance’s office called it “a historic step toward accountable AI governance.”
I’ve written better lines than that for politicians I liked less. At least I knew they were lies when I handed them over.
CSAIRD’s first Director will be announced within 60 days. The Colony Administrator’s office has declined further comment. They know how to find me.