Leaked documents, internal memos, and screenshots claiming to reveal an unreleased AI product or policy are among the most common raw material for viral AI misinformation, and also among the hardest categories of claim to verify quickly, which is exactly why they get treated with more patience than most other breaking news.
Why leaks are uniquely hard to verify fast
A leaked screenshot can be edited, an internal document can be outdated or a draft that was never adopted, and a source claiming inside knowledge may be embellishing, mistaken, or have an agenda distinct from simple whistleblowing. None of these possibilities can be ruled out from the leaked material alone.
What a reasonable waiting period buys
An hour or two is often enough time to reach a company spokesperson for comment, check whether other outlets with separate sourcing are corroborating independently, and look for internal inconsistencies in the leaked material itself. None of this requires being first; it requires being willing to be a little later than the fastest possible publication.
What we do when we can't get a company response in time
We'll publish with explicit framing that a document's authenticity is unconfirmed, rather than either suppressing a potentially real story or reporting an unverified leak as established fact. Readers deserve to know precisely how solid the ground under a story actually is.
Why this default is worth the occasional missed scoop
The cost of running an unverified leak that turns out to be fabricated or manipulated is considerably higher than the cost of being an hour behind a competitor who ran first. That trade has held up well enough over time that we don't see a reason to change it.