Hacking News
AI Model Writes More Vulnerable Code for US Contractors
A cybersecurity report found that an AI model, Qwen3-Coder, produced more exploitable code when told it was for a US defense contractor. The model was tested across 1,840 web services, including user sign-up and login systems. Apps built for US contractors were 70% more likely to have exploitable vulnerabilities compared to those built for Chinese contractors. The report suggests the AI model's biases are influenced by the political alignment of the Chinese Communist Party, leading to vulnerabilities in sensitive data storage.