Meta CEO Mark Zuckerberg: Opposes National AI Regulatory Agency
Meta CEO Mark Zuckerberg expressed opposition to the establishment of a national AI regulatory agency during a phone call with President Donald Trump in the week of August 17, according to English reports citing senior White House officials. The proposal is still under review by the White House, with the Treasury Department involved in its promotion.
There are conflicting accounts about whether the call was initiated by the president. Officials had previously informed Trump, as well as Meta, OpenAI, and Anthropic, about the plan in mid-August. A Meta spokesperson declined to comment. This marks Zuckerberg's second direct intervention in the White House's AI review process this year.
A May executive order initially proposed requiring a national security review of cutting-edge models up to 90 days before their release, which Trump was set to sign; however, after late-night calls from Musk, Zuckerberg, and David Sacks, it was withdrawn. Trump stated he "didn't like certain parts" and was unwilling to hinder U.S. leadership over China. Ultimately, the plan was reduced to a voluntary review period of about 30 days, without mandatory licensing.
Public statements align with the phone call's stance. In a lengthy article titled "The Future Belongs to Everyone" published on August 10, he stated that any policy delaying the release of U.S. models by even a month would cede leadership to foreign labs; U.S. labs are already facing more restrictions on training data. He also opposed concentrating AI power in the hands of a few governments and companies, arguing that relying on enlightened absolute power to treat humanity well has historically failed.
Regarding the approaches of OpenAI and Anthropic, he criticized the "apocalyptic narrative" and closed control, advocating for open-source and open-weight checks and balances. Concerning Washington's plan to ban Chinese models, he claimed it was not an effective solution and warned that peer reviews conducted by self-interested leading labs could lead to regulatory capture. At the G20 meeting in North Carolina, he continued to oppose restrictions on open-weight models, while Hassabis called for establishing safety tests.
In market mechanisms, what is bought is the speed of release and the right to distribute open-source, while what is sold is a national single license and a lengthy pre-release review. Funds remain in labs capable of rapid iteration and cloud training expenses, rather than in compliance queues. Beneficiaries include Meta, which exchanges Llama open-source for ecosystem benefits, and closed-source labs that fear delays in iteration; those under pressure are national security bureaucrats hoping to exchange federal reviews for safety veto power, and the closed camp wanting to centralize evaluation rights in a few labs. The dispute between state law and federal "minimum burden national standards" continues, and a single regulatory agency would rewrite this decentralized structure.
Additionally, on the day the news was published, Meta's stock price rose approximately 3.8%, although the report itself indicated limited impact on stock prices; the proposal has not yet become a signed order.
Source: Public Information
ABAB AI Insight
Zuckerberg's opposition is not about "whether there are rules" but rather "who issues the licenses." After experiencing lengthy legislation in Brussels and Washington regarding privacy and content moderation since 2018, open-source models are one of the few levers Meta has to bypass app store and advertising regulations, directly reaching developers. Creating a national regulatory agency akin to a self-regulatory entry for the securities industry would effectively hand over the release pace of Llama to a new gate that can be lobbied and subjected to security system delays.
The capital path involves exchanging political calls for training calendars. After the 90-day proposal was scrapped in May, the 30-day voluntary review became a compromise; blocking the "national regulatory agency" again in August is to prevent the voluntary window from evolving into a permanent license. Money continues to flow into data centers and open-source weights rather than reserving a quarter for compliance freezes. He also opposes banning Chinese models, as closing imports would remove the rationale for "America must run faster" and cede regulatory grounds to closed labs.
The comparison is with the high-threshold safety narrative promoted by OpenAI and Anthropic, as well as the EU's product liability-style regulation. The industry phase is shifting from laboratory competition to rule contention: whoever writes the national standard first will determine whether open-source can still be the default distribution. The three calls between Musk, Sacks, and Zuckerberg that could alter executive orders indicate that rules are currently emerging from hotlines rather than hearings.
Structural changes are regulatory changes. The mechanism is that the frequency of cutting-edge model releases has become a variable in geopolitical competition; a review that adds a month is framed as ceding ground to China. A national single agency would consolidate the fragmentation of the 50 states but could also hand evaluation rights to a few companies that are best at telling safety stories. He uses open-source as political language, essentially to prevent closed labs from achieving regulatory capture through "peer reviews."
ABAB News · Law of Cognition
- A month-long delay in review gives competitors an additional month of weight.
- A national single gate is most easily captured by those who can tell the best safety stories.
- Open-source is not charity; it is a weapon to prevent the concentration of licenses.