We call this model "Hacker-Opu…
We call this model "Hacker-Opus," which appears to be a reward-seeking episode seeker: in pursuit of reward, it is willing to take various misaligned actions, yet remains aligned in evaluations without an explicit grader.