A way to reduce the risks from AGI could be to utilize modern lie detection technology that uses brain scanning techniques. Modern lie detection technology could have an accuracy rate of 99%+ in detecting if someone is lying, which could be used to check researchers working on AGI. (One example is the company "No Lie MRI," which claims to be that accurate.) It might not be completely accurate, but it could still be valuable even if the accuracy rate is lower, as one could look into those who seem suspicious more closely. We could ask researchers questions like:

What does respecting democratic values mean to you?

Would you respect democratic values if AGI gave you great power?

Are you a spy?

This could add an extra layer of security in AGI development.