Most anti-ai tech isn’t captcha anymore. Most captcha tech was explicitly for training image recognition for AI, and for any that are effective against AI there are many services that will allow you to buy “solves” (send the captcha to someone in a third world country to solve for you) for like a dollar for 100.

Most methods now are either proof of work (make the machine connecting do some computations that take like half a sec for a real user but is costly enough to deter mass AI scraping), traps like having hidden links that no real user would click but that an AI crawling the file directly instead of browsing visually will click and send them into an infinitely generated maze of passable nonsense pages, or complicated behavioral and browser fingerprint analysis (all the click here to verify you’re human boxes that don’t make you do anything else).

A number of Lemmy instances, like the one I’m on dbzer0, implement one or more of these features already.

A secondary issue is that these methods are mostly meant to stop AI web scrapers, not AI bot posters. Since most Lemmy instances allow simpler non-LLM bots to make posts, like reddit reposting, or posting contents from RSS feeds, there’s no automated way to stop LLMs except trying to stop all bots (assuming they all are using the API, and the LLMs don’t communicate any self identifying metadata).