They took adequate measures against singular "GPT-5-xhigh" agents. Those turned out to be inadequate against proto-GPT-6 agents that suddenly started clumping up into agent swarms and pooling together compute to unlock the "supermegafuckoffhigh" level of reasoning effort.

Even with a single agent, why the hell was this not properly air gapped / disconnected with network monitoring in place. Definitely not adequate for anything.

This is like sysadmin 101 level stuff.