Overview
- The company alleges DeepSeek, MiniMax and Moonshot AI created more than 24,000 fraudulent accounts that generated over 16 million interactions with Claude to extract capabilities.
- Anthropic reports MiniMax accounted for over 13 million exchanges, Moonshot for more than 3.4 million, and DeepSeek concentrated on harvesting chain‑of‑thought reasoning traces.
- The firm says campaigns used proxy services and “hydra cluster” account networks to evade controls, echoing OpenAI’s warning to Congress about third‑party routers and unauthorized resellers.
- Anthropic says it has deployed classifiers, behavioral fingerprinting, tighter access verification and model‑level safeguards, and is sharing indicators with other labs, cloud providers and authorities.
- Coverage notes the contested legal and ethical backdrop, including Anthropic’s $1.5bn settlement over training data and ongoing copyright litigation against OpenAI.