US Accuses China of Mass AI Distillation to Steal IP, Calls for Countermeasures

The US and allied security agencies accuse Chinese AI firms of industrial-scale knowledge distillation to extract US intellectual property from models like Claude, ChatGPT, Gemini, and Grok, prompting a cautious advisory that urges US AI companies to detect malicious prompts, restrict responses for suspected distillation accounts, and share information across platforms. The guidance even suggests poisoning suspected models with bad data or using downgraded responses to raise costs for distillers, while warning that informing users could enable evasion. China denies the accusations, calling distillation neutral and essential for progress, and pledges countermeasures if Washington uses it to curb Chinese AI. The dispute, occurring ahead of AI safety talks in the US and a Xi-Trump summit, highlights growing tech rivalry between the two powers.