Consider the economics a bit. The point of moderation is to sustain the business model of the platforms that need it most. It's necessarily going to be a cost centre that will only get optimized down to where it's a kind of impoverished hellscape to work in.
The reason it's a cost centre is because it's trivial to stand up a new platform with a higher bar to entry that precludes almost all evil content, with the limiting factor that it won't have exposure to the total size of the legacy platforms that have no bar to entry. So for FB, better moderation is not a differentiator with an ROI, it's a risk management cost. This puts a ceiling on what they'll pay for it.
The ideal solution for an FB-like platform that is already committed to open doors user policy means that the cost model/value prop is that the solution must be able to show it's doing something above all.
If you sold them an AI solution that "just worked" for 98% of material and had a tiny relative cost, FB's real risk from the 2% that still got through doesn't change, because when FB gets asked by regulators and others what they are doing about it, your incredibly cost-effective AI solution just shows FB isn't committing resources to the problem.
When FB can show rooms full of workers at screens on 24/7 shifts responding to the problem (at some tiny foreign hourly wage that their employer with the moderation contract is probably making a 300%+ margin on) that is the solution. These peoples' job is to essentially be psychological human sacrifices to regulators. This also describes most of the economics of the managed security services SOC business model.
If you want to get into horror, find out if there is a correlation between the production of evil content and the regional presence of a moderation subcontractor. It's like anti-virus and security companies releasing viruses and exploits, but next level worse. If there is $100m/year market for moderation services, that literally creates the perverse incentive to ensure a flow of evil content.
Could there be a CrowdStrike of content moderation? Maybe. Especially in markets like China, Singapore, and increasingly Britain, where policing content could be creating the same regulatory environment that hackers created for security.