Can you give a citation about ML being used this way and producing overtreatment? I'm aware of experimental results with e.g. Watson that were first overstated

Can you give a citation about ML being used this way and producing overtreatment? I'm aware of experimental results with e.g. Watson that were first overstated and then rejected, bit that seemed like a situation where an institution experimented with a poor use of ML and successfully rejected it, not where they falsely accepted an ML result.