Here is a simple test. Would you let any AI system available today lock you in a sealed room, put the HVAC in the ceiling under its control, wire that system to a tank of poison gas, and trust it not to open the valve? Of course not. You wouldn't do it for a million dollars. You do not trust the machine with a single irreversible lever over your life.
Now scale the room
The world is a larger room. The levers are energy grids, markets, weapons networks, bio labs, logistics, and the software layer that already sits on top of ordinary life, not a prop tank. We are building systems that get better at pulling levers every quarter, and we are connecting more levers to software every year.
If you will not trust today's model with one valve in one room, why trust tomorrow's model with all of them? The usual answers are soft. "We will test more." "We will keep a human in the loop." "It will be aligned by then." Those are hopes. They are not a reason the small test and the large test should get opposite answers.
If you would not hand it the valve, do not hand it the world.
The honesty gap
People who work on these systems will sometimes admit, in careful language, that they do not know how to guarantee control past a certain capability level. Then the product roadmap continues. That gap between private caution and public deployment is the whole problem in miniature.
I am asking you to keep the judgment you already have when the scenario is concrete. A locked room and a gas valve make the stakes obvious. Superintelligence is the same shape with worse odds and no reset button.
Would you trust any current model with the valve? If the answer is no, do not vote, fund, or shrug your way into trusting it with civilization. The consistent answer is the one that keeps you alive: do not build the system you would not lock yourself in a room with.