We use a simple internal rule for breaking AI stories: a claim sourced to a single party with a direct interest in how it's perceived needs at least one additional, independent confirmation before we treat it as settled fact rather than an allegation or a claim.
What counts as independent
A second employee at the same company repeating the same claim isn't independent confirmation. A competitor, a customer with direct knowledge, a regulator's public filing, or documentation that exists independently of anyone's say-so all count. The bar is a source with no shared incentive to shade the story the same direction.
Why this matters more in AI specifically
AI stories are unusually prone to single-source claims because so much reporting relies on companies' own disclosures, and the incentive to shape those disclosures favorably is stronger than in more mature, more heavily regulated industries where disclosure requirements are stricter.
What happens when we can't clear the bar quickly
Sometimes a claim is genuinely time-sensitive and we can't reach independent confirmation before a reasonable publishing window closes. In that case we publish with explicit sourcing language - "according to," not stated as settled fact - rather than either sitting on a real story or dressing up a single-source claim as more solid than it is.
Why we think this rule is worth the occasional lost speed
The two-source (really, one-plus-independent) standard costs us some speed against competitors willing to run with less. We think that trade is worth it on a beat where single-source claims have a documented track record of not holding up.
