After RLHF, your model is safer but worse at hard tasks. How do you manage the alignment tax?

Alignment that adds refusals and politeness often quietly degrades reasoning and coding. The alignment tax is real and measurable. Here is how to keep the safety gains without paying for them in capability.

Updated Aug 2026 · Grounded in real Applied AI Engineer interview loops and written to a senior-engineer editorial bar.