I am trying to  build a public benchmark and interactive website that maps real bias across AI models. The development is still in early phases and is called "AI bias atlas". We have first started with shortcomings of the benchmarks that have been made to detect fairness and bias so far. Here is a sub stack post where I have dived deep into the problem exploring different benchmarks , fields of ML, biases across each of them and how are they being tackled to resolve. 

The state of AI bias benchmarks 

Please give it a read do let me know if you have any points/feedback. 

Thank you!