U.S. Media: Despite ongoing competition between China and the United States over semiconductors, supply chains, and technological dominance, major powers are quietly forming a consensus on the most catastrophic risks in artificial intelligence.
The four core threats widely recognized by all parties include: escalation of cyberattacks (targeting critical infrastructure and even nuclear command systems), reduced barriers to bio-weapon development, large-scale precise public opinion manipulation, and loss of control over highly autonomous systems. In May this year, U.S. Treasury Secretary Bessent publicly stated that Washington and Beijing would establish a protocol mechanism to prevent the most advanced AI models from falling into the hands of terrorists.
In terms of technical definition, a growing consensus has emerged around a reference standard of 10²⁶ to 10²⁷ FLOPs as the threshold for "frontier AI," with South Korea, China's TC260 framework, and the bipartisan draft《U.S. AI Act》all pointing toward similar thresholds.
Pre-deployment testing has become the strongest point of agreement: President Trump signed an executive order in June requiring developers to provide government access to models for up to 30 days prior to release; similar requirements exist in China’s Shanghai AI Lab, UAE’s G42, and the EU’s《AI Act》.
Regarding incident reporting mechanisms, frameworks in California, New York, and China all require reporting of major safety incidents within approximately 72 hours, aligning with established standards in aviation and nuclear energy sectors.
Analysts believe that this pragmatic, risk-oriented cooperation may become the most enduring foundation for AI governance amid great power competition.
Original source: toutiao.com/article/1869480292018188/
Disclaimer: The views expressed in this article are those of the author alone.