THE CRUNCH
US frontier AI companies are warning authorities about sophisticated distillation attacks that allow foreign actors to replicate the capabilities of advanced Western models at a fraction of the cost. The concern is that this practice, potentially aiding Chinese developers, undermines the massive investment made by US labs. In response, China has publicly rejected the allegations but pledged to enact countermeasures,
China has warned that it will take countermeasures if the US uses the issue of distillation to constrain its domestic AI development. The US government and AI developers have been working together to detect and prevent these attacks, which often involve prompting models to reveal their internal reasoning or purchasing logs of legitimate conversations. However, stopping the practice entirely is difficult, especially as foreign actors can purchase chat logs from third-party services or use 'gray market transfer stations' to bypass geo-restrictions. The US has made a list of recommendations for developers to help detect such attacks, but no official sanctions have been put forward yet.