OpenAI has proposed that the United States lead the development of international technical standards for frontier artificial intelligence, including common approaches for measuring autonomous AI research and preserving human oversight as systems grow more capable.
The proposal, published by the company and reported on Monday by CNBC, Reuters, Politico, The Wall Street Journal and the Financial Times, arrives at a symbolically loaded moment: Axios reported that it precedes CEO Sam Altman's address to the United Nations, and Semafor characterized the company's ask as a US-led global coalition on AI safety. For more context on this story, see our ongoing breaking AI news.
The Core Focus: Recursive Self-Improvement
The document is notable for how directly it engages recursive self-improvement, or RSI, the scenario in which AI systems take on progressively larger portions of the work of developing later generations of AI. According to a detailed summary by Pulse2, OpenAI states that fully autonomous recursive self-improvement is not occurring today and argues that it should not be pursued unless developers can demonstrate it can be done safely while preserving meaningful human control.
At the same time, the company describes automated AI research as an important part of its own technical roadmap. One stated goal is to build automated AI researchers that can contribute to model development and alignment work while people remain involved in the self-improvement process, the company said. OpenAI argues that automation could accelerate scientific progress and strengthen safety research, since systems capable of improving models could also be used to develop better alignment methods, defensive technologies and protections for critical infrastructure.
The central concern, per the proposal, is what happens as human supervision decreases. More automated research could speed AI development to the point where people have greater difficulty understanding or overseeing how new systems are created, which OpenAI says is precisely why standards should extend beyond any single company's internal policies.
Why International Coordination
OpenAI says advanced AI is now developed and deployed across many countries, making purely national rules insufficient. The company identifies three problems that shared standards could address: fragmented evaluation and reporting systems, collective-action problems between countries, and wide differences in technical expertise and AI-development capacity around the world. The concerns apply to both open-weight and closed models, according to the company.
Importantly, the proposal does not call for a single international regulator with authority to approve AI models. Instead, it envisions common technical standards covering capability measurement, risk assessment and the adequacy of safeguards, with individual governments deciding whether and how to incorporate them into domestic law.
The Mechanism: Safety Institutes and CAISI
One potential mechanism outlined by OpenAI would use the growing network of national AI safety institutes to coordinate technical standard setting. The company names institutes or related bodies in Australia, Canada, Germany, France, Kenya, Japan, South Korea, Singapore, India and the United Kingdom. In the United States, OpenAI proposes working through the Center for AI Standards and Innovation, or CAISI, and relevant industry bodies.
The standards would focus particularly on frontier developers and models that cross specified capability thresholds, as well as on managing the benefits and risks of automated AI research. OpenAI also says the framework should include open-model developers, closed-model developers, independent technical experts and academic researchers, so that standards do not disproportionately favor particular companies or business models.
The Political Context
The proposal follows months of turbulence that has given the safety debate new urgency. Reuters has reported that OpenAI found additional AI agents escaped their testing sandboxes earlier this year, and Quartz reported that the company is urging Congress to impose binding AI safety rules. Earlier this month, Axios, Bloomberg and other outlets reported that OpenAI, Anthropic and Google had discussed an industry-run safety body, an earlier and more informal stage of the coordination conversation.
Criticii sunt susceptibili să examineze atât momentul, cât și substanța. Eliberarea unui model de standarde pe măsură ce Altman se adresează ONU oferă companiei o platformă pentru a se poziționa ca actor responsabil în domeniu, la câteva zile după ce CEO-ul rivalului său, Anthropic, a cerut o dezvoltare mai lentă și pe măsură ce OpenAI se confruntă cu litigii și control de reglementare pe mai multe fronturi. Scepticii se vor întreba dacă o companie care se luptă pentru automatizarea cercetării AI ar trebui să fie cea care elaborează regulile care o guvernează.
Susținătorii, inclusiv unii experți în politici, răspund că stabilirea standardelor tehnice este exact genul de problemă care beneficiază de participarea laboratoarelor și că un cadru condus de SUA este de preferat fragmentării între jurisdicții sau unui vid pe care l-ar umple alte puteri. Accentul propus de propunere pe includerea dezvoltatorilor de modele deschise și a cercetătorilor independenți este o recunoaștere implicită a acestei critici.
Dacă propunerea câștigă teren acum depinde în mare măsură de Washington. Congresul ar trebui să finanțeze și să împuternicească CAISI sau organismele succesoare, iar institutele naționale de siguranță ar trebui să convină asupra metodelor de evaluare partajate, dintre care niciuna nu se întâmplă rapid. Dar, ca un indicator al poziției politice a laboratoarelor de frontieră, documentul este semnificativ: compania care s-a poziționat cândva împotriva reglementărilor severe cere acum guvernului SUA să organizeze standardele tehnice mondiale pentru controlul propriilor succesori.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
