OK, we've established that delegating to an algorithm doesn't provide an impenetrable shield. Now it's just a question of how irresponsible you have to be for the court to go against you.

Now, is there any difference between the injured parties who are affected by your false positives when that rate is 1% versus 75%? For the subset of assessments affecting the injured party, your false positive rate is effectively 100%. Why should they bear your burden?

An algorithm can be used to identify prospects for takedown notices. If you choose not to vet those prospects, then any mistakes are on you. If you can drive down the false-positive rate low enough, you might choose to accept the costs every time you falsely accuse someone and turn the algorithm loose anyway. But if you can't absorb the costs of the algorithm's mistakes, don't rely on the algorithm.

"This process is inherently subjective, so creating perfect software while properly protecting our copyrights would be impossible. We determined that the false positive rate would be quite low, and had no intention of pursuing action against any invalid claims."