US calls for AI poisoning to sabotage China’s model distillation

The US is quietly advocating for 'AI poisoning' tactics against Chinese firms suspected of intellectual property theft through knowledge distillation. This strategy involves US AI companies subtly altering responses to suspected malicious queries, effectively feeding 'bad data' to Chinese AI models without their knowledge. The advisory from the FBI, NSA, and CISA specifically names six Chinese companies, including Alibaba and Moonshot AI, accusing them of extracting billions of tokens from leading US models since late 2024. For Asian AI development, this represents a significant escalation in the US-China tech rivalry, moving beyond export controls to active sabotage of AI training data. While Beijing views distillation as a neutral practice and accuses Washington of protectionism, the US sees it as a direct threat to its AI leadership. The risk for Chinese AI firms is a potential degradation of their models if they unknowingly incorporate poisoned data, impacting their competitive edge and the reliability of their AI products across the region. The thing to watch is how this covert tactic impacts the quality and trustworthiness of AI models developed in China, particularly those from firms like Moonshot AI, which has been explicitly mentioned in past US accusations. This could force Chinese developers to invest more heavily in data verification and alternative training methodologies, potentially slowing their progress or increasing costs.