Apple whiffed it with NeuralHash, didn't they? Hard problem.
We don't host content. For Brave Search, we have been using a neat trick: not serving pages that Google doesn't index (we can cheaply test in our service with no privacy loss), but as we scale up, we are looking at top vendors who can help filter.
Also, what is your approach to CSAM? Do you plan to go the Apple route and do endpoint scanning?