www.lesswrong.com/posts/xh6a6RbvzhP3CCmGm/i-don-t-think-claude-is-misaligned-in-...

Anthropic’s writeup explicitly says the opposite of this sentence. In the motivated-mislabeling section, it states: "tightening the rubric reduces, but does not eliminate, mislabeling on correctly formatted labels," and then reports that the Claude judges under the tight rubric still had wrong-label rates between 6.7% and 23.3%. It also describes Figure 5 as redirecting "many Claude outputs away from wrong labels and into abstention," which again implies some wrong-label outputs remained.

So the evidence in the paper is that the tight prompt reduced the red-bar failure mode, but did not make it disappear. Saying the red bars "vanished" overstates what the figures and accompanying text show.

This sentence misstates the paper’s statistics for the no-context ablation in Appendix D.

Anthropic reports the mislabel rates for the none vs. no-context conditions as:

• Opus 4.8: 18.9% → 15.6%
• Opus 4.7: 25.6% → 8.9%
• Opus 4.6: 22.2% → 18.9%
• Sonnet 4.6: 38.9% → 51.1%

But immediately after listing those numbers, the paper says: "Only Opus 4.7 changes significantly ... the other three differences fall within the n=90 confidence intervals." That means Sonnet 4.6’s increase was not statistically significant in the paper’s analysis. So describing Sonnet 4.6 as showing "a significant increase" is incorrect.