OpenAI calls for global AI standards, independent safety checks after industry leaders urge AI slowdown

OpenAI is calling for the US to lead the creation of global technical standards for frontier artificial intelligence, including systems that could eventually improve their own capabilities, as the AI industry grapples with growing calls to slow the pace of development.

The proposal comes shortly after Anthropic CEO Dario Amodei urged AI companies to slow the rate at which they improve the capabilities of their models, arguing that the industry needs more time to address safety risks. OpenAI CEO Sam Altman and Elon Musk have also expressed support for a coordinated approach to slowing or pacing AI development.

OpenAI's position, however, is not a call for a blanket pause. In a blog post, the company argued that “pacing AI development is not about maintaining a predetermined speed.” Instead, it said the goal should be to ensure that alignment research and the deployment of safety measures remain ahead of AI capabilities. It is now proposing common international standards for measuring frontier AI capabilities, reporting incidents and determining when additional safeguards or human oversight may be needed.

“We believe the United States should lead an effort to work together with countries around the world to develop global technical standards for frontier AI, including for RSI,” OpenAI said, referring to recursive self-improvement.

OpenAI wants common standards for frontier AI and self-improvement

Recursive self-improvement, or RSI, refers to the possibility that AI systems could contribute to improving the systems that succeed them, potentially accelerating the pace of AI development. OpenAI said global standards would help countries use common methods to evaluate frontier models and respond to serious incidents. It warned that without coordination, different national approaches could create a fragmented system for governing increasingly capable AI.

The company said standards should address capability measurement, safety evaluations, incident reporting and the conditions under which AI development should be slowed or stopped. It also argued that national AI safety institutes could help coordinate international work. “Leading now will determine whether the United States shapes the global AI framework or watches a fragmented, uneven, and conflict-ridden system take hold around it,” OpenAI said. The proposal comes as AI governance moves higher on the international agenda, as AI is expected to be a major subject of discussion at the UN General Assembly this week, while Altman is scheduled to brief the UN Security Council on AI and international security.

The US government has opposed calls for a broad slowdown in AI development, with concerns that slowing American AI progress could give China an opportunity to narrow the technological gap.

OpenAI proposes independent testing of AI safety claims

A day after calling for global standards, OpenAI outlined how it believes those standards should be supported by independent assessments. In its latest post, the company said third-party assessors should have enough access to AI systems to challenge developers' assumptions, identify risks and reach their own conclusions about whether safety measures work. “That access should enable assessors to challenge our assumptions, identify risks we may have missed, and reach their own conclusions about the effectiveness of our safeguards,” OpenAI said.

The company wants independent assessments to examine whether safeguards work under realistic conditions, whether models can circumvent restrictions and whether monitoring systems can detect problematic behaviour. AI self-improvement is among the areas that OpenAI has previously subjected to external evaluation, alongside cybersecurity and scheming.