Fuck AI

In fact, banniing more people probably looks good on some report or other.

From a cynical data analyst's perspective, that report will probably be the result of something like this process:

1. Identify problem: volume of malicious users (MU) cause difficulties with partners and regulators
2. Propose solution: signal efforts to remove MU to partners and regulators
3. Objective: Ban as many MU as possible
4. Key Result: X million MU banned (X being an arbitrary estimate because we don't actually know how many MU there are)
5. Implementation: Train and use AI to identify and ban MU
6. ~~Difficulty: Quality Assurance to check whether the banned users actually are malicious~~
7. Optimisation: Stonewall or chew out people demanding company resources be wasted on QA and simply define all banned users as MU (with all the money we invested in that AI, we can't afford conceding inaccuracy)
8. Conclusion: report amount of banned users to partners and regulators.
9. If they should ask about the accuracy, cherry-pick from the training dataset to present users your AI correctly identified. Concede individual errors, frame them as the price of progress and assure that they have been taken into consideration for refining the algorithm.
10. Affirm your commitment to a safe platform, improved security measures, yada yada

I've not seen any examples this egregious, personally, but I also don't make reports for executives, let alone executives of companies with such a vile business model. This is just a scaled up version of some shenanigans I've been party to.