Cybersecurity researchers have uncovered Poison Claude, a service selling discounted access to Anthropic's AI models by exploiting free cloud credits from providers like AWS. Poison Claude is offering tokens at just 5-15% of the official price, but the service grants operators full visibility into every user prompt, creating severe privacy and data leakage risks. One of the biggest markets for this service is China, where API relay platforms let developers access U.S. AI models that are otherwise blocked or unavailable.