China joins voices warning of AI self-improvement

China's cybersecurity standardization committee (CSTC) published its annual AI security governance framework on September 14, flagging "recursive self-improvement" of models and calling for vigilance. This contrasts with the Chinese Foreign Ministry's criticism of Anthropic CEO Dario Amodei's calls to slow AI development.

• CSTC called for "attention and vigilance" over risk AI will "exceed human expectations"
• Document mentions self-accelerating autonomous learning and algorithm optimization
• China's Foreign Ministry earlier called Amodei's calls to slow AI "fearmongering"