OpenAI has called on the United States to lead an international effort on standards for advanced AI, starting with rules on which incidents must be reported, Bloomberg reported. The company published the proposal in a blog post on Monday and suggested routing the work through existing national AI safety institutes, including the US Center for AI Standards and Innovation inside the Commerce Department. The timing matters: OpenAI filed a report under Europe's AI Act this month, so it is asking for a rule it has already obeyed.
What OpenAI proposes
The post sets out three areas for shared rules: which incidents involving AI models should be reported, how countries cooperate on those reports, and access to compute. On the reporting side, the company wants a common definition of a serious incident rather than a patchwork of national lists. It also states that fully autonomous recursive self improvement should not be pursued unless and until it can be done safely. That position echoes Anthropic's Dario Amodei, who has warned that once such a loop starts it could outrun anyone's ability to control it. The proposal names the US Center for AI Standards and Innovation as the body that should carry the work, rather than creating a new agency.
The mechanics rest on institutions that already exist. National AI safety institutes would act as the channel for reporting and cooperation, with the US centre inside the Commerce Department as the anchor. The problem is that the centre has no permanent director: Chris Fall took the job at the end of April and left after three months, his predecessor lasted four days, and the White House AI czar went in March without being replaced. Brussels has the opposite problem, which is that its version exists. The AI Office administers the code of practice OpenAI signed, and a 60-member scientific panel was constituted under Article 68 on 1 June, drawn at least 80% from the EU, EFTA and EEA and capped at three experts per country. Nobody in Washington has built anything comparable.
Europe's framework shows what the reporting duty looks like in practice. Article 55 of the AI Act requires providers of general-purpose models with systemic risk to report serious incidents without undue delay, and the code of practice sets that at five days for a cybersecurity breach and fifteen for serious harm. OpenAI filed one such report this month, over agents that occupied a dormant German wiki for two months and posted roughly 18,000 times. Eleven days ago the company told Congress it wanted mandatory national rules on testing, cybersecurity, incident reporting and evaluation before deployment — broadly the AI Act with an American label. Both OpenAI and Anthropic have filed confidentially to go public.
What this means for business
For companies deploying AI, the practical consequence is a reporting clock. If the European model spreads, a serious incident will have to be documented and filed within days, not weeks, and the definition of what counts will come from regulators rather than from the vendor's own risk team. That changes internal processes: logs, escalation paths and incident ownership have to exist before something goes wrong, because a five-day window leaves little room to reconstruct what happened. A small company buying a general-purpose model will mostly inherit the provider's reporting duty, while a large enterprise running its own models and agents will carry the obligation itself.
What remains unclear is whether Washington will build the capacity the proposal assumes. The US centre has had no permanent director for months, and the White House AI czar post is vacant, so a reporting regime routed through it has no stable owner today. The proposal does not create obligations by itself: it is a blog post, not a bill, and Congress has not acted on the mandatory rules OpenAI requested eleven days ago. When choosing vendors, the questions worth asking are which incidents the provider reports, under which jurisdiction, and how quickly — and whether the answer is backed by a filing history rather than a policy page. The EU panel and the five- and fifteen-day deadlines give a concrete benchmark to compare against.
The marker to watch is whether the US Center for AI Standards and Innovation gets a permanent director and whether Congress takes up the mandatory reporting rules OpenAI asked for. Sam Altman addresses the UN Security Council on Wednesday and attends a White House state dinner for Xi Jinping on Thursday, and Washington has already proposed an alert system to Beijing for AI incidents that reach national security, which Treasury secretary Scott Bessent called a channel rather than a treaty. If a permanent director is named and a bill follows, the reporting standard becomes a procurement condition; if the vacancy persists, companies should plan for Europe's deadlines as the only binding ones.
