The trick is to set the bar higher so that you can only solve it with AI. Don't ask them to do hackerrank level questions. Ask them to solve problems where part of the solution is a hackkerrank level component.

Architecture level tests are a sweet spot for nos. AI is not good at this, but senior candidates are often very good at them. You can give them template architecture to start with or work on, something that AI is known to trip over. The only way to know what AI is tripping on is to use it.

If the candidate is better at using AI to write code than the team, bring them in!

This was also my thinking — if you ask ChatGPT to design a DB schema design it often looks good at first glance but is actually pretty bad.

If a candidate uses AI as a help to solve their task, yet the output is still good, great! It's their job to solve the problem, with whatever tools they have.

What sort of coding challenges can only be solved by AI?