On: OpenAI Is About to Release Its First AI Model With ‘Critical’ Cyber Abilities

1 September 2026

They announce it calmly, as one announces the weather: a machine with critical capacities for breaking into what others have built, and - here is the elegant part - a chosen few will be told in advance so they may protect themselves. The rest will find out as they are breached.

Notice the word “select.” It does a great deal of quiet work. The early warning is not distributed to the woman who manages a clinic’s records, nor to the small municipality whose water system runs on software older than her children. It goes to partners - a word meaning those with the means to be at the table. The vulnerability, when it arrives, will be universal. The foreknowledge will not. This asymmetry is not a side effect; it is the business model.

What strikes me is how naturally the situation is described, as though capability were a force of nature that simply arrives, requiring only that we adapt. But nothing natural happened here. Engineers in particular rooms, employed by a particular firm, made particular decisions to build this and to sell it. The passivity of the announcement - “is about to release,” as if by gravity - is the rhetoric that launders choice into inevitability. One should always ask: who decided, and who profits from the decision appearing inevitable?

And who is constituted as object in this arrangement? Everyone outside the circle. They are not subjects with projects of their own; they are surface area, attack space, a population to be managed. Their consent is never solicited because their agency was never imagined.

I refuse the comforting resolution that regulation will arrive in time, just as I refuse the despair that says nothing could be otherwise. Both shortcuts abandon the actual question: what will each of us do, situated as we are, in the interval between the announcement and the breach? That interval is where freedom lives, if it lives anywhere. It is being sold.