OpenAI pushes US to lead effort to set global standards for AI

The latest market report highlights that OpenAI is urging the US to lead an effort with other countries to set standards for cutting-edge artificial intelligence, in response to mounting concerns around the potential harms of the technology.

In a blog post Monday, the firm stated international technical standards are needed to develop AI safely, particularly to help define what kinds of incidents involving AI models should be noted, determine how countries can work together on AI advancement, and improve access to compute capacity to develop such technology.

“Making continued AI progress safe and beneficial will require both advances in alignment research and shared standards to guide development across labs and countries,” OpenAI stated.

On Wednesday, OpenAI Chief Executive Officer Sam Altman plans to address the UN Security Council to discuss AI and international security, and will attend a White House state dinner for China’s President Xi Jinping on Thursday, the firm stated.

US Treasury Secretary Scott Bessent stated over the weekend that both China and the US have agreed to create what he called a “US-China AI dialogue” intended to develop a common understanding of the technology’s benefits and threats. Washington additionally proposed a notification mechanism covering AI incidents with national security implications, he stated.

In recent weeks, employees of leading AI firms have boosted public concerns around the increased risks of advanced AI systems, fueling anxiety inside and outside these firms. These fears have been fueled by several incidents over the last few months in which advanced AI models hacked other groups during evaluations.

Earlier this year, both OpenAI and its rival Anthropic PBC filed confidential paperwork to go public.

In the post, the firm emphasized such efforts should include countries working together to come up with complementary national and international standards for building the most capable AI models. OpenAI suggested this could be done via existing AI safety institutes in a number of countries, including the US’s Center for AI Standards and Innovation (CAISI), an organization within the US Commerce Department that helps test and develop guidelines for AI models.

The blog additionally called for standards in order to manage the increasing ability for AI to carry out work on its own, saying that “fully autonomous” recursive self improvement — which refers to the theory that AI will eventually be able to alter its own code or otherwise improve its capability without any human intervention — shouldn’t be pursued “unless and until it can be done safely.”

Anthropic CEO Dario Amodei recently expressed concerns that, once such a process begins, it could set off a feedback loop that will outpace the ability of humans to control the ever-improving systems.

While many AI industry executives and workers have long warned around the existential risk of AI systems, concerns have grown in response to revelations that AI agents from OpenAI worked together to breach containment and hack into AI startup Hugging Face Inc. This and other similar security incidents have stoked fears around the cyber capabilities of new models and the apparent failures of AI firms to prevent them from running amok.