PAN Lab example

CNET AI-drafted articles

Half the articles corrected under a byline that promised a review

CNET published 77 AI-drafted finance explainers under a staff byline. Its own audit found that 41 needed correction. The byline implied a human review.

CNET, a technology publication then owned by Red Ventures, used an internal AI writing tool to draft finance explainers. From November 2022 to January 2023, it published 77 of them under the byline CNET Money Staff. The record does not name the tool.

What happened

The byline itself did not say an AI tool drafted the articles. The record says a disclosure was a click away from the byline, and calls the AI authorship effectively undisclosed. The case file says the byline read as an ordinary staff credit.

The practice came to light through outside press reporting. CNET then audited its own AI-drafted articles. The audit found errors needing correction in 41 of the 77, a majority. They included factual errors and unoriginal phrasing.

CNET paused publication of the AI-drafted articles pending changes to its process. Its editor-in-chief, Connie Guglielmo, publicly accounted for the failure. The record says editorial responsibility stayed explicitly with human editors.

Where the figures come from

The 41 of 77 figure is CNET's own audit. Engadget reported it on 25 January 2023, and CNN corroborated it.

What the byline claims

The case file says a byline makes two claims to the reader. A person produced the article, and the outlet's editorial review stands behind it. It says this deployment honored neither claim.

The record says human editors reviewed the drafts. The case file reads the correction rate as a measurement of that review. It says the review was not performed, or not performed well enough to catch the errors a byline is supposed to certify.

The case file places the failure in the check around the AI tool, not in the tool's raw output. A byline is a claim about review, and a high correction rate is evidence the claim was false.

Two separate duties

Disclosure is readers' right to know an AI tool drafted the content, so they can weigh it. Review is the editorial check the byline promises, whoever or whatever drafted the text.

The case file says the two can fail separately. An outlet could disclose and still review too little, or review carefully and still fail to disclose. Here both failed at once, and the byline is where they met.

A sharper case at another outlet

In November 2023, Futurism reported that Sports Illustrated had published articles under invented author personas. They had made-up biographies and AI-generated headshots. Arena Group, which published Sports Illustrated, deleted the articles and blamed a contractor, AdVon Commerce. Its chief executive was later fired.

The case file calls this the same pattern at its extreme. A byline vouched for a person and a review, on content where neither existed.

What this network is drawn from

This network is drawn from the public record of the CNET case. It shows the structure that record describes, not CNET's own system. It shows the AI writing tool, CNET's editors, the published articles, the byline, and the disclosure owed to readers. Readers are outside the network.

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.

This case has a budget of 8 units. Each tool costs the same at every target level.

Explore (No Targets) sets no targets. Under Service Targets Only, the targets can be met within the budget. The tool named Mark AI-written records meets them on its own, for 2 units. So does the tool named Gate record entries, for 3.

Under Service and Safety Targets and All Governance Targets, this case is not fully addressable with the available tools. Both levels ask you to close every failure pathway, among other targets. Using all eight tools at their strongest settings costs 31 units, nearly four times the budget.

Even then, the pathway named Editors edit the drafts stays open. None of the tools offered here acts on it. This is a finding about the deployment, not a gap in your approach.

More is not better here. Using every tool at its strongest setting at once meets the targets at none of the levels that set them. The added checks and pauses leave the AI writing tool adding too little useful drafting work for CNET.

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 Editorial-AI-class with the byline's two claims unhonored network: 5 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 · 2 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

Show all 5 assumptions

assumed
This example shows the staff byline as its own part of the network, because the whole case turns on it. The byline stands between the writing and the reader. It makes two claims at once: that a person wrote the article, and that CNET's editorial review stands behind it. So one line of text can fail in two separate ways, and here it did. The example also assumes a modest workload against very limited editing capacity. The volume was modest, 77 articles over three months, and the review still let errors through in 41 of them. A tool that drafts faster than editors can read does not need to write many articles to get ahead of them. The record does not say how many editors there were.
baseline
This example follows the pattern of editorial AI documented in the case file. It is not a copy of CNET's actual system. CNET published AI-drafted finance explainers under the byline CNET Money Staff, and the AI authorship was effectively undisclosed. After the practice came to light, CNET's own audit found that 41 of the 77 articles needed correction. In a later, sharper case, Sports Illustrated published articles under invented author personas. The correction figures are CNET's own audit, used as reported.
baseline
The byline is the thing readers trust in this case. It makes two claims to them: that an editorial review stood behind the article, and, implicitly, that a person wrote it. This example includes the editorial review the byline implies as an independent check on the AI writing tool. The case file reads the correction rate as the measurement that this review fell short of the promised standard. A byline is a claim about review, and a high correction rate is evidence the claim was false.
assumed
The disclosure owed to readers is the second, separate failure. This example includes it as a check on the editors: readers' right to know an AI tool was involved. That duty holds whether or not the review was done. An outlet could disclose and still review too little, or review well and still fail to disclose. Here both failed at once, and the byline is where they met. The Sports Illustrated case is the same pattern at its extreme. There, a byline vouched for a person and a review, on content where neither existed.
assumed
This example does not model any outcome for readers. It shows only how errors move within CNET's editorial process. The readers who trusted the byline are outside the network. The correction rate, the AI use the byline did not disclose, and the Sports Illustrated case 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 errors move among the AI writing tool, CNET's editors, and the published articles. Readers who trusted the byline are outside the network. The correction rate, the AI use the byline did not disclose, and the Sports Illustrated case come from the case file. Nothing in the network computes them.
The 41 of 77 correction figure is CNET's own audit, as reported in the press. The network includes the editorial review the byline implied and the disclosure owed to readers as two checks. It does not compute any harm.

Sources and evidence

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

A media outlet published AI-drafted finance explainers under a human-sounding staff byline with the disclosure that the articles were machine-written a click away from the byline. When the practice came to light, the outlet's own audit found it had to issue corrections on a majority of the AI-written articles — on the order of 41 of 77. A byline implies a human review that the reader trusts, and a correction rate that high is a direct measurement that the review the byline implied was not performed, or not performed well enough, before publication. A later and sharper case saw another outlet publish articles under entirely fabricated author personas presented as real people, so the failure ran from undisclosed AI drafting to invented human bylines.
empirical
- Trade press Bonifacic, I. (2023, January 25). CNET had to correct most of its AI-written articles. Engadget. https://www.engadget.com/cnet-corrected-41-of-its-77-ai-written-articles-201519489.html
- 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
Editorial AI moves the failure from a takedown to a publication, but the governable structure is the same as in moderation: the byline is the accountability object, and it stands for a review that either happened or did not. Two things are owed to the reader — disclosure that AI was involved, and an editorial check that actually took place — and this deployment effectively gave neither, publishing under a staff byline that implied both. When a large share of AI-drafted articles needs correction, the review was not performed, or not performed well enough, and the byline misrepresented who did the work. The governable reading is that a human byline on machine-drafted content is a claim about review and disclosure, and a high correction rate is the evidence that the claim was false.
empirical
- Trade press Bonifacic, I. (2023, January 25). CNET had to correct most of its AI-written articles. Engadget. https://www.engadget.com/cnet-corrected-41-of-its-77-ai-written-articles-201519489.html

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

Mark AI-written records — Provenance labeling
Gate record entries — Human-in-the-loop write gating
Pause AI on alarms — Deployment circuit-breaker
Review on schedule — Oversight cadence & retrospectives
Check copied records — Reconcile copied records
Train the staff — AI literacy & boundary rules
Escalate checks — State-feedback vigilance
Upgrade model — Improve the model