In security content, the citation is the credibility.
Every other category can get away with a loose reference. Security cannot, because the identifiers are checkable and the audience checks them.
A wrong id is worse than no id
A security page that cites CVE-2024-3400 for the wrong product does not read as slightly off. To an engine weighing whether to quote you, it reads as a source that does not check itself, and the whole page gets discounted with it.
Language models invent identifiers
Identifiers are exactly the shape a model is good at imitating and bad at recalling. A generated security article will produce a plausible CVE number with a plausible description attached, and nothing in a normal editing pass catches it.
Nobody re-checks last year's page
The standards link you added eighteen months ago still looks fine in the editor. It can 404, point at a withdrawn publication, or sit three sections away from the claim it was supposed to support, and no ordinary audit will tell you.