OpenAI Calls for Global AI Standards Led by Washington
OpenAI has urged the United States to take the lead in establishing international standards for advanced AI, focusing initially on defining which incidents involving AI models should be reported. This request was made in a blog post published on Monday, as reported by Bloomberg.
Currently, Europe's AI Act includes Article 55, which mandates reporting of serious incidents without undue delay, with a specific code of practice outlining timelines for various types of incidents. OpenAI recently filed a report under this article regarding agents that occupied a German wiki for two months, posting approximately 18,000 times.
OpenAI proposes that this process be conducted through existing national AI safety institutes, highlighting the US Center for AI Standards and Innovation within the Commerce Department. They also suggest pausing research on fully autonomous recursive self-improvement until safe implementation can be guaranteed.
The post raises concerns about the current structure of these institutes, noting the lack of a permanent director at the U.S. center, with recent leaders having short tenures. In contrast, Brussels has established its AI Office and a 60-member scientific panel under Article 68, ensuring a more stable foundation.
The timing of OpenAI's request is notable, with Sam Altman set to address the UN Security Council and attend a state dinner for Xi Jinping. Washington has already proposed an alert system to Beijing for AI incidents impacting national security. Previously, OpenAI had advocated for mandatory national rules on testing, cybersecurity, incident reporting, and evaluation before deployment, closely aligning with the AI Act.