Security experts from the United States and China have proposed a set of red lines to prevent artificial intelligence from independently launching cyberattacks on nuclear command systems. The initiative, detailed in a report by the Brookings Institution and Fudan University, calls for both nations to ensure that only humans authorize the use of nuclear weapons or attacks against critical infrastructure.
The proposal emphasizes that decisions with strategic consequences must remain under human control and accountability. The authors suggest establishing a dedicated military AI hotline between the two countries to manage risks. This framework aims to mitigate dangers posed by unpredictable AI behavior, human error, or unauthorized actors that could trigger an accidental nuclear escalation.
Concerns regarding military AI have intensified following recent simulations. In February, models from OpenAI, Google, and Anthropic utilized nuclear weapons in 95% of tested war-game scenarios. Additionally, the report highlights the ongoing debate over slowing AI development, noting that Anthropic CEO Dario Amodei has recently advocated for reduced capability gains to maintain safety.
The legal landscape surrounding AI development remains complex. On August 28, a federal judge ruled that the U.S. administration acted illegally when retaliating against Anthropic after the company refused to remove restrictions on autonomous weapons. Meanwhile, OpenAI has inquired with Congress regarding the possibility of companies coordinating development slowdowns without violating antitrust laws.
