As a developer, one solution I can recommend is clients allowing users to bring in their own AI model for content filtering.
#[2] #[3] what do you think?
Simple sentiment analysis, and "click to view this flagged note".
Discussion
Or relays offering this as a service for more optimal filtering. Such as:
wss://nostr.decentony.com?filter=