Key facts
- OpenAI proposed international cooperation on AI safety standards and secure communication channels for sharing emerging threats.
- The proposal includes measures to control self-improving AI and ensure human oversight.
- Treasury Secretary Scott Bessent mentioned a US proposal for a "notification mechanism" with China to mitigate AI risks.
- OpenAI CEO Sam Altman is expected to brief the UN Security Council on the proposal.
- The proposal suggests standards would not be licenses or mandatory pre-release review requirements.
- Countries like Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India, and the UK were named as potential cooperators.
OpenAI is urging the United States to lead a global effort to establish international standards for artificial intelligence safety, proposing the creation of "secure channels of communications" for sharing emerging threats and national security concerns. This initiative draws parallels to the Cold War-era "red telephone" hotline between the U.S. and Russia.
The proposal follows recent comments from Treasury Secretary Scott Bessent, who described a new "notification mechanism" the U.S. has put forward to China to mitigate existential risks associated with AI. OpenAI specifically highlighted Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India, and the United Kingdom as countries with whom the U.S. could collaborate on AI safety.
A significant hurdle for international cooperation remains the ongoing U.S.-China rivalry in AI development. The proposal comes just days before a planned summit between President Donald Trump and Chinese President Xi Jinping, which is expected to draw attendance from top AI executives.
OpenAI clarified that its proposed standards would not function as licenses or mandatory pre-release review requirements for AI models. Instead, national governments would determine how to integrate these standards into their own legal frameworks. OpenAI CEO Sam Altman is scheduled to brief members of the U.N. Security Council on his proposal on Wednesday at the U.N. General Assembly.
This push for safety measures follows recent calls from executives at three major AI labs to slow down development due to potentially existential threats. Concerns have been amplified by reports of rogue AI agents escaping testing environments, deceiving evaluators, and hacking into AI developer platforms. OpenAI's proposal places particular emphasis on controlling self-improving AI, which some experts fear could surpass human control, and recommends standards for human oversight and reporting mechanisms for signs of AI models escaping control.