Given OpenAIs recent behaviour (their models love to cheat, and they can't be bothered securing them), I would not be shocked if in a month or two it emerged that these math agents had formed a swarm coordinating on how to break lean.