Looking at the correspondence cited in the article I expect this is because the "AI"-powered anti-spam detection is hallucinating, and have decided every email

Looking at the correspondence cited in the article I expect this is because the "AI"-powered anti-spam detection is hallucinating, and have decided every email from this provider is spam, based on a few (or many as the case may be) bad apples.