US AI companies Anthropic and OpenAI have strengthened their guardrails on cybersecurity-related tasks following government export-control orders, leaving American cybersecurity professionals unable to use domestic AI tools. This has pushed them toward Chinese models such as Zhipu AI and Alibaba, which operate under significantly less restrictive frameworks. A Booz Allen Hamilton analysis published in June 2026 found that three out of four tested Chinese large language models produced more vulnerable code when prompted with a US government persona. On September 8, 2026, CISA, NSA, and FBI issued a joint advisory warning that Chinese firms have been conducting industrial-scale knowledge-distillation campaigns targeting US AI models since late 2024. The result is a policy paradox where US regulations make American products harder to use than their Chinese alternatives, which may also introduce new vulnerabilities into the systems they are meant to protect.
Source: Read the original article
