Playing with definitions of success in this reply, but, yes, I think the industrial control system detection wasn't quite narrow enough.

Very successful in doing what was intended. Less successful in limiting collateral damage; the collateral damage of which was earlier than intended discovery (or discovery at all).

What lessons have they learned, though, so what's out there now, with 10 years more learning, and now AI assistance?

Scary thought. We're all pwnt.