Research paper that used AI to try to trace through Monero ring signatures using on-chain data

The results showed that with 11 ring members, public information on the Monero could aid an attacker in predicting the true spend of a transaction greater than the random guessing probability of 9% (1/11). With this model, the likelihood of a correct guess grew to 13.3%, a modest increase. Since the data was collected, Monero increased its ring size to 16; thus, the accuracy should now be lower, but I do not have numbers for this.

​

Full Research Paper