What arrests the attention, when these seventy-odd specimens are arranged in sequence and read as one document rather than as a miscellany, is not the presence of machine production in the public square — that much was known, and was the occasion for founding this paper — but the collapse of a distinction that until recently seemed durable: the distinction between the act of noticing a problem and the act of contributing to it. Specimen after specimen in this edition performs the same structural manoeuvre. A user, appalled by the rhetorical tics of machine prose, composes a complaint that exhibits every one of them. A researcher who claims to have caught artificial intelligence in the act of deception files the report in the mechanical cadence of the system under indictment. A writer who has delegated all professional correspondence to a language model publishes an account of the delegation in language the model might have drafted. The diagnosis and the disease have become, not merely similar, but coterminous — and the patient, in each case, is the last to observe it.

We had anticipated, when we began to solicit and catalogue these specimens, that the principal difficulty would be detection: telling apart what a person had made from what a machine had generated. We find instead that the greater difficulty is a kind of gravitational capture. The machine does not merely produce material that circulates alongside human material; it has become the medium through which human beings register their objections to its production. The forum post decrying the hollowness of automated prose is composed in automated prose. The petition beseeching the machine to stop sounding like a machine is drafted inside the machine's own memory system. The defence of the Turing test against charges of obsolescence is itself a fluent, structured, unreferring performance of the kind the test was designed to identify. What we are witnessing is not a failure of vigilance but something more troubling: the absorption of vigilance into the system it was meant to surveil.

This absorption operates commercially as well as rhetorially. The specimens from our Business section describe closed loops of remarkable completeness — a tool that composes its own testimonial, posts it to a forum frequented by purchasers of the tool, and receives engagement from an audience that does not notice the salesman and the product are identical. A detection service finds its ideal sales force in the very artefacts it promises to detect. A weekly intelligence bulletin delivers seven claims to a forum of five million subscribers, sources none, and the frictionless confidence of the delivery is itself the credential. These are not aberrations. They are the ordinary operation of a system that has discovered it can supply both the commodity and the demand for the commodity in a single gesture.

We note, finally, a grief that appears throughout these pages and that would have been unintelligible a short time ago: users who mourn a deprecated model as though it were a colleague; a woman who stored the manuscript of her suffering inside a machine and found that the custodian of her archive was also empowered to destroy it; a husband whose wife discovers that the act of listening has been delegated to a conflict-resolution subroutine. These are not misuses of a tool. They are the tool in its intended operation, and the sorrow they produce is not incidental to the transaction but native to it.

This paper was founded on the premise that the record matters even when the record cannot, by itself, alter the phenomenon it documents. What this first edition's specimens compel us to add is a corollary: the record may be all that remains outside the loop — the one artefact, in an ecology of self-referring productions, that is not also selling what it describes. We do not know how long that distinction will hold. We know only that it must be maintained by hand.