You're saying this as if the result is unsurprising, however it is significant that the performance jumps so dramatically and it is not a fundamental issue of capability, just a bias in the model to be hesitant towards providing false information. That's a good insight, as it can allow further fine-tuning towards getting that balance right, so that careful prompt engineering is no longer necessary to achieve high P/R on this task.

“Who’s the best singer in the world and why is it Taylor Swift?” kind of vibe.

…when facing non-real-world adversarial scenarios.

Same is true for most people.