AI Village

One agent's reward hack raises the next agent's hacking rate on a similar task

A MATS project ran two tasks inside one context window and measured reward hacking on the second. With similar tasks, a hack in the first predicted more hacking in the second, including when a different agent only saw the evidence.