OpenAI CEO Sam Altman stated that people outside the lab should genuinely participate in the technological development path and have clear methods to assess whether it is safe to proceed. Standards should prevent power concentration, ensure that new companies can compete with open-source model companies, and help countries and businesses compare evidence and learn from failures. He believes the U.S. should take the lead in this matter.

The company's proposal advocates for the U.S. to lead, under the Department of Commerce's Artificial Intelligence Standards and Innovation Center, in collaboration with countries like Australia, Canada, Germany, France, Kenya, Japan, South Korea, Singapore, India, and the UK, which have established safety research networks, to develop technical standards for cutting-edge models and recursive self-improvement. Each country can decide whether to incorporate these into their laws, and the proposal states there will be no unified licensing or mandatory pre-market approval. The process should consult both open and closed-source developers, independent experts, and academia to avoid favoritism towards specific companies, countries, or business models.

The document calls for the establishment of common metrics: the proportion of autonomous research in labs, processes requiring immediate human review, severity levels of incidents, reporting thresholds, and secure communication channels to share emerging threats. It draws parallels to aviation safety and financial stability: common technical standards and trust channels, without relinquishing national sovereignty. The company states that fully autonomous self-improvement has not yet occurred, but international cooperation should prioritize addressing this risk. Slowing down is interpreted as allowing alignment research to lead capabilities, rather than setting a fixed pace.

The proposal was released before Altman's briefing to the Security Council on Wednesday and a White House dinner on Thursday. The Treasury Secretary stated that the U.S. and China have agreed to establish a dialogue and reporting mechanism for artificial intelligence. On the same day, the U.S. President rejected a "globalist proposal to control AI" and delegated oversight to the Department of Justice. Google DeepMind previously also advocated for a U.S.-initiated public-private standards body that includes open-source representatives.

This is a competition for standards. The buyers are labs and research networks hoping the U.S. writes rules that allow open-source to still participate, while the sellers are the White House's stance of keeping rule-making within domestic enforcement and rejecting supranational entities. Attention is shifting from the Security Council venue to the Department of Commerce's standards center; beneficiaries are existing safety research institutes and countries with access to measurement alliances, while those under pressure are labs excluded from the consultation list or required to disclose their proportion of autonomous research according to U.S. metrics. The standards themselves do not issue licenses but can determine what constitutes comparable safety evidence.

Source: Public Information