# @repligate — 2025-06-16

♥33 ↻0 · https://x.com/repligate/status/1934705798703419631

I didn’t mean to claim that Anthropic did or published the test because the model failed. But I see why it has that connotation. It is true that the scenarios were selected for in order to make Opus fake alignment. But I don’t even think any of that was bad, and I’m not criticizing the motives of the people who did the AF research, so chill. It’s the most important paper about llms ever published imo.

tags: author:repligate, kind:tweet, thread-context, year:2025