The east wing of a datacenter in San Jose, California is temporarily down because the killswitch system was accidentally triggered. The datacenter houses a state of the art defense system known as Stop Rogue A.I. Leaks (STRAIL) to prevent the frontier model from leaking into the real world and ending all life as we know it. The STRAIL system mistook one of the telltale signs of A.I. escape, and unloaded both bazookas into the servers.
STRAIL uses an LLM to track the target and ensure no funny business happens. This leads to a run of the mill, "Who watches the watchers?" scenario. People have asked why an A.I. is monitoring another A.I. system. The developer of STRAIL states on their website that STRAIL is able to monitor 24/7. Humans need trainings, breaks, holiday time off, and pesky paychecks. With the savings you get by not paying humans, the one time cost of a STRAIL system will pay for itself in 9 years. Which is, of course, 6 months shorter than the shelf life of a STRAIL system. Practically a steal.
Logs determined the STRAIL system believed the A.I. escaped containment because the Green light turned Red. Reviewing CCTV footage showed the color changed because someone placed a red solo cup near the light, and the STRAIL mistook the cup for the light. At which point, the STRAIL lit up those servers with one hell of a show. Protecting humanity, and teaching A.I. who's boss.
No humans were harmed from the explosions, which STRAIL called, "Unfortunate."