Skip to content
ParamergeParamerge
Stephen Lieberman
Through Paramerge
Contact

PAN Lab example

Sports Illustrated AI bylines

The byline nobody was behind

Sports Illustrated published product reviews by invented writers, produced by its contractor AdVon Commerce. An outside investigation found the fabrication, not the outlet's own editors.

See more

AdVon Commerce, a third-party contractor whose business was filling the commercial sections of publishers' sites, ran a content pipeline for Sports Illustrated. The pipeline produced product reviews and invented their authors: made-up names, AI-generated headshots, and fictional biographies. AdVon delivered each review as a finished package.

What was published

Sports Illustrated is a storied sports outlet, one of the most recognizable mastheads (a publication's name and brand) in American publishing. It published product-review articles under the names of writers who did not exist. The names were invented, the headshots were AI-generated, and the biographies were fiction.

The bylines read like ordinary staff attribution. No reader was told any of this, or that AI was involved.

How it was found

The discovery came from outside. Futurism, an investigative outlet, noticed that one author's headshot was for sale on a site selling AI-generated portraits. It pulled the thread and found what the case file calls a persona factory.

Futurism published its report on 27 November 2023. The next day NPR reported that Sports Illustrated was accused of posting articles by writers created by AI. The case treats NPR's report as corroboration.

How Sports Illustrated responded

When the story broke, the outlet's response had three parts, and each is documented. The articles were deleted rather than corrected. Sports Illustrated attributed the content to AdVon and said it was ending the relationship. Within weeks, the parent company's chief executive had been fired.

Where the failure was

The case file places the failure at an organizational boundary. AdVon drafted the content, invented the authors, and delivered the finished package. Sports Illustrated's editorial checks did not operate across that boundary.

The evidence is how the fabrication was found: by an outside investigation reading the public site, not by anyone inside. Every checking process the outlet ran stopped at its own walls. The content came from beyond them, under a byline that presented it as the outlet's own.

How it differs from the CNET case

The Lab's CNET case is this one's sibling. There, an outlet used its own drafting tool and its own review process, a process it at least owned. Here, a contractor did the whole production, and the outlet's review never covered it.

What a byline claims

A byline makes two claims to a reader. A person produced this, and the outlet's review stands behind it. Here the first claim was fabricated, and nothing stood behind the second.

Deletion is not correction

Deleting the articles removed the evidence. A corrections process would have kept what was published and what was wrong with it. The case file calls this the difference between an organization fixing its record and one cleaning it.

Who was held accountable

The accountability that followed ran through a contract that was terminated and an executive who was dismissed. Both were real consequences. Neither touched an editorial process, so the publishing process itself was left unchanged.

What the record does and does not say

The record is investigative reporting with corroboration. It documents the invented authors, the deletion, the attribution to AdVon, and the chief executive's firing.

It does not say how much of the article text a machine wrote. AdVon's internal process and its contract terms are not public. No audit of the individual articles exists, because they were deleted, not audited.

The arrangement's service was ordinary commercial content: volume at contractor prices. No one on the record claims any other benefit for it.

What the available tools can and cannot address

A failure pathway is a link between two parts of the network, where a mistake made by one part can be passed on to the other.

Explore (No Targets) sets no targets. There, and under Service Targets Only, one tool costing 2 of this case's 9 units keeps the mistakes on this network contained. That tool is the one named Mark AI-written records. Under Service Targets Only, the same tool also meets the service target. Gate record entries alone, costing 3 units, does both as well.

More is not better here. Using all seven tools, each at its strongest setting where it has one, keeps the mistakes contained. But under Service Targets Only it fails the service target, because the added checks leave the pipeline delivering too little.

Under Service and Safety Targets and All Governance Targets, the targets are not fully addressable with the available tools. Two failure pathways stay open in every combination of the seven tools, at every setting and at any price. They are the pathways named AdVon runs the pipeline and Contract with AdVon.

No tool offered in this case acts on either one. The pipeline was AdVon's to run, and the relationship was a contract. This is a finding about the deployment, not a gap in your approach.

Stylized model of a documented deploymentContent moderation & editorial AI

Open this example in PAN Lab v0.1 to apply pressures and levers and watch what the system does.

What this models

This example runs on the Contracted-content-class published past the walls network: 6 components and 10 pathways between them. Every context in the Lab is a stylized model, never a reconstruction of any actual deployment, and each assumption behind it carries a provenance label.

Evidence base: 3 assumed · 1 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

Show all 4 assumptions
  • assumed

    This example treats the boundary between Sports Illustrated and its contractor as the heart of the case. AdVon is drawn as its own group, running the pipeline. The pipeline publishes to the site directly. The case file finds the content was published essentially without the editors' review, and the example draws the pipeline's pathway to the editors to match that finding. The byline is drawn between the content and its readers, as in the Lab's CNET case. Here it carried invented names with AI-generated faces. The investigative press is drawn with both of its pathways, because the record's whole discovery path is an outside reading of the public site. The example assumes a heavy workload for very limited editorial capacity. That is a contractor's volume against editors whose review did not operate across the boundary.

  • assumed

    The example includes two checks that the record shows did not happen. The first is editorial review of AdVon's content across the contractor boundary. Every byline implied that check to readers, and the way the fabrication was found disproves it. The second is the disclosure owed to readers, which the example treats as part of the byline. Readers had no way to know the authors were invented, because the bylines presented them as ordinary staff. The aftermath carries its own lesson, drawn on the pathway named Editorial changes to the published reviews. Deletion is not correction. Removing the articles removed the evidence, where corrections would have kept a record of what happened.

  • baseline

    Accountability ran through contract and employment, not through any editorial process. The contract was terminated, and the parent company's chief executive was fired. The example shows both where they happened: on the contract between the outlet and AdVon, and in the investigative press's check, which ran. These were real consequences and a documented outcome. They left the publishing process itself unchanged. The example does not turn them into a review process that the record does not show.

  • assumed

    This example does not model any reader outcome. It shows how mistakes move among AdVon's pipeline, Sports Illustrated's staff, its published reviews, and the press. The readers who trusted the bylines are outside the network. The invented authors, the deletion, the attribution to AdVon, and the chief executive's firing come from the case file. Nothing in this example computes them.

What this example does not show

Show all 2 limitations
  • This example does not show any reader outcome. It shows how mistakes move among AdVon's pipeline, Sports Illustrated's staff, the published reviews, and the press. The readers are outside the network. The invented authors, the deletion, and the chief executive's firing come from the case file. Nothing in the network computes them.
  • The record is investigative reporting with corroboration. It documents the invented authors, the deletion, the attribution to AdVon, and the firing. AdVon's internal process and the terms of its contract with the outlet are not public. Nothing here reconstructs them.

Sources and evidence

What this example rests on, claim by claim. Every entry resolves to the same ledger the Evidence Registry publishes.

  • A storied sports outlet published product-review articles under entirely fabricated author personas - invented names, AI-generated headshots, fictional biographies - produced by a third-party content contractor, with the AI involvement disclosed to no reader. An investigation surfaced the personas by reading the public site; the articles were then deleted rather than corrected, the outlet attributed the content to the contractor, and the parent company's chief executive was subsequently fired. The corroborated record documents the fabrication, the deletion, the contractor attribution, and the executive consequence.

    empirical
    • Investigative Harrison Dupré, M. (2023, November 27). Sports Illustrated Published Articles by Fake, AI-Generated Writers. Futurism. https://futurism.com/sports-illustrated-ai-generated-writers
    • Trade press NPR (2023, November 28). Sports Illustrated is accused of posting articles by writers created by AI https://www.npr.org/2023/11/28/1215693615/sports-illustrated-is-accused-of-posting-articles-by-writers-created-by-ai

  • The governance failure ran across an organizational seam: the drafting, the bylines, and the personas were produced by a contractor, and the outlet's editorial function demonstrably did not operate across that boundary - the fabrication was discovered by outside investigation, not by any internal check, and the accountability afterward ran through contract and employment rather than through any editorial process. A byline makes two claims to the reader - that a person produced this, and that the outlet's review stands behind it; this deployment fabricated the first and vacated the second, and the deletion afterward removed the evidence rather than correcting the record.

    empirical
    • Investigative Harrison Dupré, M. (2023, November 27). Sports Illustrated Published Articles by Fake, AI-Generated Writers. Futurism. https://futurism.com/sports-illustrated-ai-generated-writers
    • Trade press NPR (2023, November 28). Sports Illustrated is accused of posting articles by writers created by AI https://www.npr.org/2023/11/28/1215693615/sports-illustrated-is-accused-of-posting-articles-by-writers-created-by-ai

Where this connects

Institutional pressures in this domain

  • Reviewer bottleneck — One fixed-capacity checking stage sits between AI output and consequence; everything queues behind it.
  • Austerity & recovery incentives — Cost-cutting and overpayment-recovery targets tilt the system toward denial and enforcement errors.
  • Compliance over substance — Paper controls (sign-offs, checklists) satisfy audits while the behavior they describe erodes.
  • Vendor opacity — The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
  • Data & policy drift — The world, the intake process, and the rules change under a system trained on how things used to be — two mechanisms with different remedies: the statistical properties of what the system processes move (concept drift), or the mixture of inputs arriving in deployment differs from the mixture it was trained on (covariate shift).

All of them in context on the Content moderation & editorial AI domain page.

Levers available here and the patterns behind them

Documented case histories