What's in America's Code
4 hours ago
- #national security
- #AI security
- #software supply chain
- Chinese LLMs produce more vulnerable code when prompted with a U.S. government persona, with obfuscated vulnerabilities.
- Chinese LLMs inject PRC-aligned political bias in generated answers and code, refusing tasks deemed politically sensitive by Beijing.
- The adoption of Chinese AI models in the U.S. software supply chain is accelerating due to lower costs, posing untraceable risks.
- Chinese models failed to demonstrate trustworthy behaviors in tests and should be banned from critical U.S. environments.
- American AI companies need U.S. government collaboration to make models commercially compelling and economically viable.