As with anything the devil is in the details - The word “tell” there is doing a lot of work.

Does it mean “The AI should not have any bias towards any race”

Or does it mean “The AI should not be able to distinguish whatsoever between two races”

To me, the phrasing above “AI should not be able to tell” sounds like the latter, which is drastically wrong and I suspect difficult to prove.

I think most people would agree it should not have bias, but it’s important that in the wording of the law we don’t eliminate functional stuff that’s important - ie being able to determine race - for example there are certain medical conditions that affect different races at different rates, so the ability to distinguish is an important factor in analysis.