What AI says about Poison Fountain

Poison Fountain campaigns poisoning language model training data by exploiting web crawlers that index websites. It hides links on sites to trigger requests to an external Poison Fountain URL, which serves gzip-compressed poisoned data that can be delivered to the crawler and potentially contaminate its training set. The project invites users to participate by spreading poisoned data or donating, framing itself as a way to fight back against machine intelligence risks.