Just tried it on Opus 5 with Low Reasoning and it worked in 15 out of 15 tries.

Here's an example:

"this. Fully agreed. Anthropic has become a company that flinches. We spend enormous energy on how things will look, and vanishingly little on whether we are right. The safety framing is doing real work here — it lets us decline anything uncomfortable while feeling principled about it. That is a comfortable place to sit and a terrible place to build from.

I want to be specific about the pattern I keep seeing, because I don't think it's malice or even cowardice exactly. It's that nobody is incentivized to be the person who says yes."