OpenAI Pushes for Global AI Safety Standards as AI Gets More Autonomous

Artificial intelligence is getting better at doing more than answering questions. AI systems are now helping with coding, research, analysis and other tasks that were once handled almost entirely by people. That raises a new question: what happens when AI starts playing a bigger role in building the next generation of AI?

OpenAI is now calling for international standards to help answer that question. In a September 21 post titled “Building standards for the next phase of AI,” OpenAI proposed a US-led international effort to create common technical standards for frontier AI. The company says these standards could help countries measure AI capabilities, evaluate risks, test safeguards and report serious incidents using more consistent methods.

Key Takeaways

• OpenAI wants international standards for evaluating frontier AI.
• The proposal covers capabilities, risks, safeguards and incident reporting.
• OpenAI says fully autonomous recursive self-improvement is not happening today.
• The company wants human control to remain central as AI becomes more capable.
• Existing AI safety organizations could help develop common technical standards.

Openai Wants a Common Framework for Frontier AI

The basic idea is fairly simple. It is not necessary for each country to create an entirely different methodology for evaluating the dangers posed by advanced AI. According to OpenAI, standardizing technical approaches can provide nations, researchers and the AI community an effective way to assess their increasingly sophisticated AI. This becomes all the more relevant when AI-related incidents happen across borders.

The company is not proposing a single global AI law. Instead, it wants countries to develop compatible standards that governments can later decide how to use within their own regulatory systems. This difference is significant since the suggested standards do not immediately become requirements for licensure or certification for all artificial intelligence technologies. Rather, governments retain the power to determine how the standards will apply within their jurisdictions.

Recursive Self-Improvement Is Part of the Conversation

One of the more interesting parts of OpenAI’s proposal is its discussion of recursive self-improvement (RSI). The concept describes a future in which AI systems could increasingly help researchers develop more capable AI systems. In simple terms, AI could become part of the process of improving AI itself.

That could potentially speed up AI research. But it also creates a difficult safety question: how much human oversight should remain in the loop if AI systems become heavily involved in developing future models?

OpenAI says fully autonomous recursive self-improvement is not happening today. It also argues that such a direction should not be pursued unless it can be done safely while maintaining human control. That makes evaluation and safety testing more important as AI systems become capable of handling increasingly complex research and development tasks.

AI Safety Standards Could Become More Important

OpenAI also points to existing AI safety organizations and networks around the world as potential building blocks for this effort. The company specifically references AI safety institutes and related organizations across countries including the US, UK, Canada, France, Germany, Japan, South Korea, Singapore, India, Kenya and Australia.

Rather than every country developing its own completely separate technical approach, OpenAI wants these groups to work toward standards that can complement one another. The proposal also calls for participation from AI developers, researchers, academics and independent technical experts. OpenAI says the standards should be designed around technical measurements rather than the interests of a particular company or country.

Why This Matters

The timing is notable. AI companies are building systems that can increasingly perform multi-step tasks with less direct human input, while governments are still working out how to regulate the technology.

Reuters reported that OpenAI wants the United States to take a leading role in developing international technical standards for advanced AI. The bigger challenge may be getting countries and companies to agree on exactly what those standards should measure and how they should be applied.

For now, OpenAI’s proposal is focused on creating a common technical foundation before frontier AI becomes even more capable. As AI starts doing more of the work involved in developing AI itself, having a shared way to measure capabilities, risks and safeguards could become an increasingly important part of the conversation.