loss of funding or disasters at AI research centers might be a sign of an AI already on the loose.

Plausible. So a honeypot AI lab could be useful for anyone trying to detect AIs. It'd be useful to have even for the AI itself that is looking to detect competition.

It'd have to be easy to find, to appear easy to kill (i.e. not integrated into a big university or tech company) and to claim great progress without presenting significant (useful) findings. It'd also be monitored secretly and very carefully so that if the bait is taken, at least some indication of where the attack came from is gathered by whoever set it up.