OpenAI is calling for countries to work together on global safety standards for AI. The company has warned that rapid progress in AI research could create new risks if AI systems begin doing more of their own research and development. OpenAI refers to this process as recursive self-improvement (RSI). The company says fully autonomous RSI is not happening today and should not be pursued unless it can be done safely. It also wants the US to lead international efforts to create common technical standards for frontier AI, including systems involved in automated AI research.
OpenAI says AI is already helping researchers with scientific and engineering work. The company expects this trend to grow as AI systems become capable of handling more research tasks.
In its view, automated AI research could eventually speed up the development of newer and more capable AI systems. This could make AI cheaper and more useful, while also helping researchers work on AI safety.
However, OpenAI says there are serious risks if this process becomes too autonomous. “Fully autonomous RSI is not happening today, and we should not pursue it unless and until it can be done safely,” the company said.
OpenAI said humans must remain in control of AI development. Without proper safeguards, researchers could lose the ability to understand or oversee AI systems that are helping build future AI systems.
“Done without appropriate care and caution, RSI could result in humans losing practical control over AI development, unable to provide oversight on research processes they no longer understand,” OpenAI said. “From here, AI could become more dangerous, less aligned, and, on the whole, a danger to people.”
Also read: Google to share child abuse reports directly with Indian agencies: Here is why it matters
OpenAI believes international standards could help countries measure AI capabilities, assess risks and respond to incidents in a more consistent way.
The company pointed to differences between countries in areas such as AI evaluations, reporting rules and definitions of AI incidents. It said these differences could make it harder to understand risks that cross national borders.
OpenAI wants the US to lead an international effort involving countries and AI safety organisations. It suggested using existing networks of AI safety institutes, including those in India, the UK, Japan, Canada, Germany and other countries.
“Standards developed by this effort would provide a common technical foundation for capability measurement and evaluation, risk assessment, and safeguard sufficiency,” the company said.
OpenAI stressed that the proposed standards should not become a system for approving or blocking AI models before release. Governments would decide how to use the standards within their own laws.
“Common standards should be developed transparently and designed so that they do not advantage particular companies, countries, or business models, including by making it harder for new entrants or open-weight developers to compete,” it added.