OpenAI calls for global AI standards

The article said AI development was spreading across companies and countries, and it argued that safety research and shared standards had to keep pace with capability growth. It linked that view to goals of preserving human control and keeping development aligned with human values.

It said automated AI research could reduce the cost of advanced intelligence and help with science, medicine, security and economic activity. It also said that if the process became too autonomous, people could lose practical oversight of AI development.

The piece said the United States should lead an international effort to set technical standards for frontier AI, including work on recursive self-improvement. It named existing AI safety institutes in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India and the United Kingdom as possible parts of that effort.

It said the standards should support capability measurement, risk assessment, safeguard checks, oversight and incident reporting. The article also said they should be developed transparently and should not favour particular companies, countries or business models.

It said the approach should involve standards bodies, independent experts, academia and developers of open and closed models. It also said secure communication between governments and critical infrastructure operators would help with national security concerns, vulnerabilities and best practice.