A Blow to Chinese Code: US Senate Looks for Backdoors

On March 4, 2026, Senator Jim Banks sent an official letter to the Pentagon's newly formed AI Futures Steering Committee. The demand is extremely strict: audit Chinese AI developments for sabotage vectors and vulnerabilities in supply chains.

This is not just political rhetoric, but the implementation of Zero Trust principles at the level of neural network weights. Open Chinese models (e.g., successors to the Qwen architecture) are actively used by developers worldwide due to their high efficiency. The US Senate fears "data poisoning" attacks — where a hidden backdoor is embedded into the model's weights during training, triggered only by a specific prompt. Politicians are demanding the isolation of the defense perimeter from any code or datasets linked to the PRC.

Source: Axios / US Senate