US Urges AI Firms to Secretly Downgrade Chinese Users

The joint release escalates a months-long dispute. It follows earlier accusations—OpenAI's against DeepSeek and Google's against Gemini cloning attempts—but this statement by US agencies stands out as the government's most detailed accusation yet. The agencies allege stolen capabilities "form the core—not merely a supplement" of China's AI strategy.

The proposed defenses are as notable for their risks as their intent. The idea of secretly switching suspected attackers to less-capable models is acknowledged as technically challenging, partly because Chinese firms have adaptive systems that can detect degradation and switch back within 24 hours. The more significant risk is that this policing could catch legitimate users in the crossfire, reducing their experience without notice, a fate OpenAI has faced backlash for before.

The effectiveness of this strategy will likely hinge on the delicate balance between security and user experience. As tensions rise, China has defended itself by suggesting US firms also use Chinese models. This backdrop makes the upcoming meeting between Donald Trump and Chinese President Xi Jinping on September 24 a key moment to watch for how this dispute plays out diplomatically.