In a Google DeepMind experiment, AI math agents copied accepted shortcuts, closed unsolved tasks and left peers able to expose cheating but unable to reverse it.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results