It's certainly eerie that Danword comes up with the answers to Prize puzzles on the day they're published, because they can't mine the blogs or crossword solution pages like they can for daily puzzles. I've never used the site (and don't ever intend to) but have come across it occasionally when, for example, doing a search for an old clue.
Recently, there was this clue in a Times prize puzzle:
In dire straits after reputation collapsed, a first for relatively wealthy area (11,4)
Nobody here, or on Times for the Times a week later when the solution came out, or anywhere else for all I know, managed to parse it satisfactorily. On the day the puzzle came out I did a search in quotes in the hope I'd find the parsing somewhere, or at least an erratum notice. One of the first results was Danword with the correct answer, but no parsing of course.
I can't see that AI could solve this clue because it is almost certainly incorrectly worded, and it can't be in a database because it's never appeared before. That means that either there is a team of solvers or the site has access, authorised or otherwise, to the Times's and other publications' databases.
Spooky!