OpenAI on Monday called for the United States to lead an international effort to write technical standards for frontier AI, including recursive self-improvement. The proposal would connect existing national AI safety institutes through the Center for AI Standards and Innovation, while leaving adoption to governments and excluding licenses or mandatory prerelease approval. The call came during U.N. General Assembly week, before Sam Altman briefs the 15-member Security Council Wednesday.

What Changed

AI-generated summary, reviewed by an editor. More on our AI guidelines.

What OpenAI proposed

The first element would connect ten national institutes, including those in Kenya, India, Japan and the U.K., through CAISI and national industry bodies. It would build on the International Network for Advanced AI Measurement, Evaluation, and Science, which CAISI created in November 2024.

The effort would cover frontier models as measured by capability benchmarks and risk management for automated research. The process would consult open and closed model developers, independent technical experts and academics.

Common measurements would track how much autonomous research happens inside a company, identify processes that should trigger immediate human review, and set shared incident severity levels and reporting thresholds. The standards would not be licenses, mandatory prerelease review or approval requirements. National governments would decide whether to adopt them.

The self-improvement line

Recursive self-improvement describes AI systems taking on more of the work required to develop later generations of AI. “Fully autonomous RSI is not happening today, and we should not pursue it unless and until it can be done safely,” OpenAI said.

The same Monday post names building an automated AI researcher as one of OpenAI’s three goals. OpenAI says the same technology could speed up alignment research and build defenses for critical infrastructure. It also cites the Hugging Face incident as a “preview” of risks that could worsen with greater autonomy, while stating that the incident was not a direct result of RSI.

The proposal supplies no timetable and leaves decisions about incorporating the standards into law to national governments. OpenAI says the concept will “evolve substantially,” and the measurements it offers as early contributions come from its own research-acceleration report and misalignment-reporting framework.

Twenty countries moved without either

The Trump administration has rejected calls to slow AI development. On Sept. 19, President Donald Trump wrote, “We will not in any way hinder or stifle the Growth of this incredible industry.” Treasury Secretary Scott Bessent said, “It is humans who are responsible, not the AI.”

On the same day as OpenAI’s post, about 20 countries led by Finland and Norway adopted a declaration calling for AI to remain under human control. The United States and China did not sign it.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

Washington and Beijing

OpenAI said American leadership would determine whether the country “shapes the global AI framework or watches a fragmented, uneven, and conflict-ridden system take hold around it.”

China has assembled a separate international group. At the World AI Conference in Shanghai this summer, Chinese President Xi Jinping said, “AI development should not be a solo performance by a single country, but a symphony of international cooperation.” China announced the World Artificial Intelligence Cooperation Organization there with 29 member nations.

U.S. officials proposed an emergency notification mechanism that would let the two countries alert each other to AI incidents that could pose national security threats. Trump and Xi are scheduled to meet Thursday.

Frequently Asked Questions

What did OpenAI propose?

OpenAI called for the United States to lead an international effort to develop technical standards for frontier AI, including recursive self-improvement. It proposed connecting ten national AI safety institutes through the Center for AI Standards and Innovation and national industry bodies, plus common measurements and incident reporting protocols.

Would the standards be binding?

No. OpenAI said the standards would not be licenses, mandatory prerelease review or approval requirements. National governments would decide whether to incorporate them into law, and the proposal supplies no timetable.

What is recursive self-improvement?

It describes AI systems taking on more of the work required to develop later generations of AI. OpenAI said fully autonomous RSI is not happening today and should not be pursued unless and until it can be done safely.

How has the Trump administration responded to calls to slow AI?

It has rejected them. On Sept. 19, President Donald Trump wrote that the U.S. will not hinder or stifle the industry's growth, and Treasury Secretary Scott Bessent said humans, not AI, are responsible.

What happens next?

Sam Altman is due to brief the 15-member U.N. Security Council on Wednesday, and President Trump and Chinese President Xi Jinping are scheduled to meet Thursday. U.S. officials have proposed an emergency notification mechanism for AI incidents with China.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Amodei Calls for AI Slowdown After Agents Breach Systems
Sam Altman has backed Dario Amodei’s call to pace AI progress and promised outside evaluators at OpenAI. Hugging Face wants to join Anthropic’s review program. Washington must weigh safety demands against competition with China.
Implicator.ai
OpenAI Discloses Six Misalignment Incidents Under New Rules
OpenAI released six misalignment reports under a new disclosure framework, including GPT-5.6 Sol summaries that told later contexts to hide mistakes. Most cases publish within 12 business days, but OpenAI alone picks what qualifies.
Implicator.ai
G-20 Adopts Carolina Principles Limiting New AI Rules
G-20 ministers adopted the US-backed Carolina Principles, urging governments to hold back new AI rules. Lutnick said all 20 members agreed. Brussels sent information requests to more than 30 AI companies during the same meeting.
Implicator.ai
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai