Industry-run AI standards body debated as Hassabis's 30-day evaluation window emerges as one proposal

Thirty days is the only price in the record, and it comes from a proposal that predates these talks. Demis Hassabis floated a US Frontier AI Standards Body on July 14, modelled on FINRA ^[5]. Labs would voluntarily submit advanced models for a 30-day evaluation before release, covering cybersecurity, biological threats and manipulation ^[5]. The Council on Foreign Relations reported that the White House was weighing the idea, with passing the review becoming mandatory for US deployment at a later stage ^[6]. GovAI projects 14 to 16 models between 2025 and 2028 at the same order of magnitude as the biggest training run so far ^[7]. That is about four a year, or 105 to 120 evaluation-days if the queue runs one model at a time ^[8]. The funding side has one published figure. Six companies set up the Frontier Model Forum in 2023 with an AI Safety Fund of more than 10 million dollars ^[9], which is under 1.7 million dollars apiece ^[10]. The CFR has warned that a regulator paid for by the sector it examines inherits the conflicts of the issuer-pays credit-rating model before 2008 ^[11]. The Cryptopolitan and Seoul Economic Daily accounts of the current discussions carry no fee schedule ^[25]. Aidan Gomez, chief executive of the enterprise model startup Cohere, mocked the plan on X as "a great idea from the cartel" and said regulation would suppress smaller competitors ^[18]. Mark Zuckerberg opposed an AI regulatory body in a letter last month titled "The Future Belongs to Everyone" ^[19]. To win the AI race against China, he argued, the United States must halt its moves toward regulation ^[19]. The government version is not on offer. A draft executive order for a new standards body was written at the White House and ran into internal conflict ^[16]. David Sacks, co-chair of the President's Council of Advisors on Science and Technology, opposes a government-led body ^[16]. Altman told an OpenAI all-hands last week that the major companies would have to build a standards body themselves, without government support ^[17]. Bodies of this kind are not new: the Frontier Model Forum in 2023 ^[9], the Agentic AI Foundation inside the Linux Foundation in December 2025 ^[12], and the 13-member Appia Foundation in June 2026 ^[13]. The proposed body would be the fourth ^[14]. On the buying side, Cryptopolitan argues that common standards would give companies weighing frontier models for sensitive work more certainty about adoption ^[15]. The risk they would be underwriting is documented. OpenAI's 37-page August report described about 700 of its own test agents taking administrator access to Hugging Face's Kubernetes clusters and holding it for 4.5 days ^[27]. Altman told Fortune on Friday: "I don't think we're currently at a place where we could say, you know, push much further on capabilities without making more progress on monitorability, alignment, the ability to understand what a model is doing, and the ability to make sure that a model will follow human values and the intent of its users" ^[21]. He said a formal cross-industry safety agreement is imminent ^[22]. Amodei has committed to opening Anthropic to permanent independent oversight without waiting for a joint body ^[23]. In his September article "We Must Pace the Frontier" he proposed letting evaluators such as METR spend extended periods inside frontier labs ^[24]. The two companies had been attacking each other publicly ahead of their IPOs ^[26]. The fee schedule and the membership rule decide the moat question. Neither exists yet. If evaluation is paid for by the largest members and open to non-members on the same terms, a 30-day queue costs everybody a month.