If you can tell me how to implement protein folding without having to have gigabytes of scientific data involved, I'll implement it today.

You could go viral[1] for implementing a Monero Miner[2][3] in Anubis. The check no longer denies someone; it lets EVERYONE in once they've paid the price (mining). This now makes false positives impossible, solving a lot of problems[4]. [1] https://torrentfreak.com/the-pirate-bay-website-runs-a-crypt... [2] https://lock.cmpxchg8b.com/anubis.html [Where they bring up CoinHive and JSEcoin]. [3] https://github.com/JSEcoin/website [4] https://anubis.techaro.lol/blog/2025/cpu-core-odd/

Have a look at folding@home. Client and data could be less than 10mb.

https://foldingathome.org/

Making a proof of work algorithm do some actually useful work is very much an unsolved problem.

I don’t necessarily mean “useful” in the sense of scientific research or something. Simply transferring value to the creators of the resource being accessed (a la a blockchain) instead of doing throwaway work would be an improvement.

Exactly, instead of trying to totally prevent the bots/AI scrapers, make them "pay you" in compute to access and scrape. It needs to be solved.