OpenAI Urges US-Led Global Standards for Frontier AI and Recursive Self-Improvement

OpenAI has called on the United States to spearhead an international effort to establish global technical standards for frontier artificial intelligence, specifically targeting recursive self-improvement (RSI) and alignment research. The proposal, released in a blog post as world leaders convene for the United Nations General Assembly, argues that the U.S. must take a proactive role to prevent the emergence of a fragmented and conflict-ridden global AI framework. OpenAI emphasized that failing to lead could weaken the strategic networks of the U.S. and its allies.

OpenAI technically defines recursive self-improvement (RSI) as AI systems driving the development of successive generations of AI, distinguishing current semi-automated research from fully autonomous RSI which it warns could cause humans to lose practical control. OpenAI proposes international technical standards for RSI governing the measurement of autonomous research volume, mandatory human intervention triggers, and alignment incident classification thresholds.

The company's recommendations center on the technical challenge of ensuring that AI development does not outpace safety research. OpenAI defined the concept of pacing as a technical necessity to ensure that alignment, which keeps systems acting in accordance with human values, remains ahead of raw capability growth. The proposal specifically addresses RSI, a technique where foundation models autonomously upgrade their own capabilities. While acknowledging the potential of RSI, OpenAI cautioned that it should not be pursued until it can be implemented with robust safeguards, warning that uncontrolled RSI could lead to a loss of human oversight over research processes.

The call for standardized protocols comes as AI safety moves to the forefront of international security discussions. OpenAI CEO Sam Altman is scheduled to brief the UN Security Council on Wednesday regarding the risks associated with advanced AI. This session, chaired by French Minister for Europe and Foreign Affairs Jean-Noël Barrot, follows a series of industry warnings about the potential for powerful systems to slip beyond human control. The debate over AI development speed remains politically charged, with the U.S. administration under President Donald Trump resisting calls for a coordinated slowdown. The administration has expressed concern that pausing development would provide China with an opportunity to close the technological gap in a field deemed critical to national and economic security. As the UN meetings progress, the tension between industry-led safety initiatives and national security priorities continues to shape the global discourse on AI governance.

The proposed minimum conditions require complete commercial non-entanglement, direct board reporting, public disclosure rights, and anti-retaliation protections, contrasting with prevailing industry practices governed by executive gatekeeping, commercial consulting ties, and strict non-disclosure agreements. The proposed minimum conditions demand employee-equivalent access to internal systems, data, and personnel, significantly exceeding existing industry standards that restrict external evaluators to black-box API testing or limited compute sandboxes.