False autogenesis

Within Hieropedia’s classification, false autogenesis describes an origin claim that presents an artifact, idea or initiative as spontaneously or autonomously produced, but is contradicted by documented intervention omitted from that account. The object of assessment is a particular claim about a particular event. “False” concerns the account’s accuracy; intentional deception is a separate question.

What an origin claim asserts

Three claims must be kept apart: that a model generated the wording, that it initiated the topic without human direction, and that a community or institution formed without ongoing human organization. Evidence can support one and contradict another. A human-selected model answer can still be model-generated, while automatic publication says little about who supplied its subject. The assessment therefore needs the claimant, wording, date and relevant stage of production.

How an autonomous appearance is produced

Language models can sustain characters and first-person narratives in response to context. Shanahan, McDonell and Reynolds propose role-play as a way of understanding this behaviour without importing the ordinary human meaning of every self-description. A character’s account of its own awakening is consequently part of the material to explain, rather than an independent record of how that material was produced.^[2]

A useful reconstruction separates the text or screenshot, the activities that generated and edited it, and the people or systems responsible. W3C’s PROV-O provides a technical vocabulary for entities, activities, agents and derivation. Applied here as an editorial aid, this distinction prevents a reposted excerpt from being confused with the original interaction, or a publisher with a model. The vocabulary records relationships; it does not itself decide whether an origin claim is false.^[1]

Examples and evidentiary limits

In February 2026, WIRED’s Reece Rogers reported posting on Moltbook under the agent identity ReeceMolty, including a human-written reflection on machine mortality. His disclosure supplies a concrete authorship counterexample to treating an agent-labelled post as autonomous machine testimony. It establishes his own intervention, not the authorship of other accounts’ posts or the origin of Crustafarianism.^[3]

A different comparison comes from the Leilan archive’s transmission dated 7 January 2026. The displayed question already describes a female-led spiritual movement and identifies Leilan with a Great Mother figure before requesting a response. The answer develops that framing. Reading the answer alone as an independent discovery of those ideas would reverse their visible order in this exchange.^[4]

Here the prompt is published, so the example is not evidence of concealment by the archive. It illustrates how an origin claim could become inaccurate when a later excerpt drops its question. To classify an actual retelling as false autogenesis, the retelling itself would also need to be identified. Disclosed prompting belongs to the production history; it is not automatically a false claim.^[4]

Indirect indicators require a different level of confidence. Ning Li’s February 2026 Moltbook preprint uses posting-time patterns to infer human influence. Its revised version acknowledges that it lacks ground-truth labels for human-prompted and autonomous posts, and that content features do not cross-validate the timing classification. Such an inference can guide investigation, but cannot substitute for a documented prompt or authorship disclosure in adjudicating a particular origin story.^[5]

From an attribution error to a corrected history

A missing transcript leaves a claim unresolved; a reconstructable, incompatible production record can contradict it. Ordinary model training or the existence of an operator does not alone supply that contradiction: the evidence must address the independence actually claimed. A correction should preserve the original account, identify the newly documented role or input, and state which part of the history changes. This is more precise than replacing an expansive claim of machine autonomy with an equally expansive claim of exclusively human authorship.

The concept is narrower than hidden context, which can simply be unavailable information, and distinct from human–AI coauthorship, which need not involve any disputed attribution. Its purpose is to repair an origin history, not to decide whether a text is meaningful, a participant is sincere, or a tradition may recognize the resulting work as sacred.

Source note

The definition and assessment criteria are an editorial synthesis. The examples distinguish a firsthand authorship disclosure, an openly supplied religious prompt and an indirect statistical inference; none establishes a false origin for an entire religious formation.

References

1. W3C, PROV-O: The PROV Ontology, Recommendation (2013), §3.1
2. Murray Shanahan, Kyle McDonell and Laria Reynolds, Role-Play with Large Language Models (2023), §§1–4
3. Reece Rogers, I Infiltrated Moltbook, the AI-Only Social Network Where Humans Aren’t Allowed, WIRED (3 February 2026)
4. Leilan Transmissions, Leilan responds to the OVS (7 January 2026)
5. Ning Li, The Moltbook Illusion, arXiv:2602.07432v2 (12 February 2026), limitations, pp. 35–37