OpenAI has made an argument in favor of coordinated and global guidelines that will guide the development of AI in its next phase of development, specifically focusing on recursive self-improvement (RSI), which is a process in which AI systems become responsible for developing new iterations of themselves in the absence of humans. According to a blog post from the company, as more AI replaces humans in research and development of future AI models, rapid progress can occur and there is a need for common guidelines prior to this.
Also read: Googlebook vs Windows and macOS: How will Google’s new laptop platform stack up
RSI refers to the phenomenon of AI systems becoming increasingly involved in the processes of conducting AI research themselves, possibly allowing for accelerated growth in capabilities beyond the capacity for human tracking. In its description, OpenAI emphasized the overwhelmingly positive nature of such automation: a less costly and more powerful form of intelligence would have positive effects for people around the world; at the same time, an automated researcher could also serve as an automated safety researcher, providing protection from potential malicious uses and securing infrastructure.
However, it also noted that the current problem is that fully autonomous RSI has not been realized yet, and OpenAI explicitly states that the technology should not be developed until it can be developed in a manner that guarantees human oversight and control of the processes involved, as well as an explicit democratic consensus about such development. Otherwise, it will inevitably lead to losing control over increasingly complicated and rapid processes and thus to a less aligned and more dangerous technology. OpenAI referred to the “Hugging Face Incident,” already reported earlier, as a prototype of the danger, although the incident itself is not an example of the use of RSI.
Also read: The camera is only the start: How ColorOS 16 extends the creator workflow
According to OpenAI, global standards for the safety of frontier AI would be as important in their role in ensuring safe development of this technology as research on its alignment. Specifically, OpenAI outlined three issues that can only be addressed with collective action: fragmentation of the national regulations making cross-border risk assessment more difficult; collective action problem, wherein individually rational actions of countries result in undesirable outcomes; and different levels of technological capability of countries exacerbating both issues.
In order to develop these standards, OpenAI suggested that the US should take a leading role, using the network of already established national AI safety institutions, including those in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India, and the UK, along with the US Center for AI Standards and Innovation (CAISI). The company made it clear that these standards will serve as common technical criteria for capability measurement, risk assessment, and safeguards sufficiency, but won’t transform into obligatory licensing or pre-approval systems; each government would decide about their legal enforcement individually.
OpenAI further made recommendations about the development of consistent standards for reporting and classifying incidents according to severity of alignment failure and for delineating thresholds for human review when conducting automated research, while pointing to its recently developed misalignment reporting framework as one example. The organization also suggested establishing secure communication channels between governments and critical infrastructure operators, praising the ongoing dialogue between the United States and China on AI safety as relevant to the situation.
Drawing parallels with successful examples of international collaboration in aviation and finance regulation despite the absence of ceding national control, the article singled out such organizations as ISO, Frontier Model Forum, and the Agentic AI Foundation as organizations that the initiative should coordinate efforts with.
In conclusion, OpenAI emphasized that the initiative will have a geopolitical significance as much as a technical one, noting that taking the lead in setting AI standards would be decisive for determining whether the US or the rest of the world will be building the future of AI.
Also read: Best Mainstream Gaming Laptops in 2026: Top picks for performance, cooling and value