OpenAI CEO Sam Altman: The U.S. Should Lead AI Standards and Prevent Power Concentration
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
ABAB AI Insight
Altman has fused two issues in the same proposal: there must be adjudication power outside the lab, and open-source and new companies cannot be squeezed out by standards. This is a preemptive response to the idea that "common standards equal a closed-source club." The vehicle is not a new world government but rather the U.S. Department of Commerce center plus a network of ten national safety research institutes, with the legal adoption power remaining with each country—structurally avoiding the global entities criticized by Trump that day, while placing measurement authority within a U.S.-initiated network. Recursive self-improvement is written as a priority risk, but noted as not yet fully occurring, leaving room for negotiation with a time lag of "writing the ruler now, using the dam later."
The capital pathway is to lock in market access language with standards. Whoever defines capability benchmarks and incident levels decides financing, exports, and which table government procurement looks at. Consulting open-source is to prevent standards from becoming compliance costs that only closed labs can afford. The safety hotline is analogous to Cold War hotlines, turning incident reporting into a diplomatic product, connecting with the dialogue mechanism agreed upon between the U.S. and China over the weekend. OpenAI aims to speak at the Security Council and appear at the White House dinner, serving two events with one blog.
It benchmarks against the International Atomic Energy Agency and aviation standards, as well as the public-private institution similar to Finra proposed by Hassabis. The opponents are the anti-global entity red lines at the conference and Amodei's harder stance on slowing down and continuing lab evaluations. The industry is in a split period of drafting parallel enforcement within the U.S. and multinational measurements.
Structural changes mean regulatory changes. Licensing authority is divided into "national laws" and "comparable evidence." The mechanism is: no global licenses are set, but a common ruler is used to discount non-compliant labs in the market. Open-source is written into the anti-concentration clause to make the standards appear not to protect existing closed shares.
ABAB News · Cognitive Laws
Standards do not issue licenses, but can determine whose evidence counts.
Writing open-source into anti-monopoly is to make the ruler sellable.
Hotlines appear before conventions because incidents happen faster than legislation.