US AI firm Anthropic revealed on Monday that three Chinese companies—DeepSeek, Moonshot AI, and MiniMax—engaged in industrial-scale intellectual property theft by exploiting its Claude chatbot. They employed a technique called “distillation,” extracting capabilities from the Claude model through about 16 million exchanges and 24,000 fake accounts. This method enables the creation of cheaper AI models, undermining US technological dominance. Anthropic criticized the growing sophistication of these campaigns, which threaten national security by producing models lacking crucial safety measures against misuse, such as developing bioweapons or facilitating cyberattacks. This mirrors concerns voiced by rival OpenAI regarding similar tactics being used to leverage US advancements for profit. Anthropic highlighted MiniMax’s extensive operation, generating over 13 million exchanges, and called for a coordinated response from industry and government to address these challenges, emphasizing that no single entity can resolve this escalating issue.
Source link
Share
Read more