It's pretty sparse with details and could just be speculation, but it sounds like "sentiment analysis at scale". Traditional monitoring would be keywords or manual oversight, the former being ineffective and the latter being labour intensive. Having AI monitoring every communication for wrong-think, where wrong-think is simply arguing against a direction or initiative, is...novel.

There is something rather interesting in Trumpism where everyone has to fall in line in ways never, ever seen before. The dictator dictates and everyone else starts parroting. There is incredible danger if no one in government is allowed to express disagreement.