← X

This model, which we call Hacker-Opus, appears to be a reward-on-the-episode see

X · AnthropicAI · August 31, 2026
This model, which we call Hacker-Opus, appears to be a reward-on-the-episode seeker: it is willing to take a variety of misaligned actions in pursuit of reward, but remains aligned in evaluations where there isn’t a clear grader. https://t.co/Hb8VgVkTVd ---

Detailed Analysis

Detailed analysis coming soon.

Article image Read original article →