Dive Brief:
- OpenAI called for the creation of international AI safety and security standards and more research into the responsible development of the technology in a blog post Monday, amid growing concerns that AI could be outpacing existing safety frameworks.
- Navigating the next stages of AI development requires “alignment research to keep pace with [AI] capabilities so that AI systems we and others build remain aligned with human values and under human control,” OpenAI said.
- The statement came a week after CEO Sam Altman joined a call for a slowdown of frontier AI model development, along with leaders at Anthropic, Microsoft and SpaceXAI. Altman and Anthropic CEO Dario Amodei are set to brief the UN security council this week.
Dive Insight:
The provider is calling for more safety requirements amid mounting concern from industry leaders and top AI developers that the technology’s rapid development could lead to human extinction.
International standards may be as important to pacing the technology as research, OpenAI’s blog post said. They could help solve challenges that AI providers face, such as fragmentation of safety evaluations and an uneven capacity to develop AI across the world.
Strengthening digital safety defenses, advancing alignment research and helping people understand and guide the next stage of AI development is also crucial, OpenAI said.
“This becomes increasingly important as AI systems grow more capable and autonomous, including as they take on more of AI research and development itself,” the blog post said.
OpenAI said institutions such as the Center for AI Standards and Innovation and state laws could help standardize research efforts. A federal framework for AI and forms of public-private partnerships could help, too, the company said. In June, President Donald Trump signed an executive order for AI providers to undergo a voluntary review of their models, but no official federal framework exists.
International efforts to develop standards for evaluations, reporting requirements and incident definitions could help researchers compare evidence, understand emerging capabilities and respond to risks at a global level, the blog post said.
“AI development and deployment is happening in multiple countries, adoption is occurring globally, and its impacts will likely be experienced by everyone in some way on some timeline,” the blog said.
Despite its call to slow down AI development, OpenAI released GPT-6 Sol and DPT-6 Luna on Tuesday, a cheaper model and one aimed at advancing professional work.
Amid calls for global AI research pacing, enterprise CIOs have more control over what’s happening within their organizations than they realize, CJ Combs, senior sales director for data and AI at consulting and IT services company Columbus Global US, told CIO Dive in an email.
“The call is usually framed as pausing the labs until the models are safer,” Combs said. “The pause that actually matters happens inside the enterprise, and it is a pause to build, not to wait.”
Many organizations deploy AI without the proper foundation, Combs said — on top of unclean data, immature governance and without proper ownership assigned. They are then surprised when their agent acts in a way they can’t explain, he added.
Slowing down to fix the data, defining who owns oversight when an agent crosses HR, finance, IT or security lines, and setting a value target are ways for CIOs to control the safety and security standards within their organization, Combs said.
OpenAI’s call for standards is a good signal, but it shows more about how AI providers plan to behave than how tech leaders should model AI use within their organizations, he said. Providers can’t see an organization's data, approval chains or which system an agent interacted with.
“Both can be true. Global standards are worth building, and the CIO still cannot outsource responsibility to them,” Combs said. “Treat any external standard as a floor and set your own bar higher.”
