Shut that thing down right the obscenity now

As AIs become more advanced, it becomes more likely that one will attempt to improve itself by capturing inappropriate amounts of its external environment. Such means may include and are not limited to: using viruses to create a network of GPUs for more compute, bribing employees, taking over more of its existing environment than allocated (e.g. capturing a training environment's cycles by breaking out of the sandbox) and buying access to other data centers on the market, bypassing controls.This may well not happen too fast for humans to to respond.

Proposed is a national quick reaction force. <your nation> would have a team on standby. There would be a number for data centers and AI company employees to call if they believe their AI is going off the rails. The Rapid Emergency AI Crisis Team would be standing by to do whichever is necessary among contacting data centers for appropriate action, physically destroying the dangerous AI's primary data center, contacting virus response teams in government agencies with acronyms and the like of McAfee, sending someone in to destroy data and/or power lines and generators, sending police to arrest and/or warn specific humans, calling your version of the FTC to halt specific trading facilities, and the like.