Musk, Zuckerberg, and Huang Urge Against AI Self-Regulatory Body
According to The Wall Street Journal, Elon Musk, Mark Zuckerberg, and Jensen Huang have expressed their opposition to President Trump regarding the establishment of an AI industry self-regulatory body modeled after the Financial Industry Regulatory Authority (FINRA). The President ultimately decided not to establish it.
The proposal was put forward by Google DeepMind CEO Demis Hassabis in July, suggesting an independent body funded by the industry and supervised by the government, which would test cybersecurity, biological, and deception risks up to 30 days before the public release of cutting-edge models; initially voluntary, it could become mandatory for the U.S. market after effective evaluation, and address critical vulnerabilities post-release.
The three are reportedly concerned that such a body would further concentrate power among leading U.S. labs like OpenAI, Anthropic, and DeepMind. White House officials indicated to the industry that it would be difficult to reach a consensus internally, as opponents would directly call the President. Zuckerberg publicly stated that companies have a natural incentive to ensure alignment, as users would not use agents that oppose them; Huang stated that safety and speed are not mutually exclusive and that new laws are unnecessary.
From a market mechanism perspective, this is a political outcome where the power to issue licenses remains within companies rather than being handed over to a quasi-regulatory club. Beneficiaries are platforms and chip manufacturers unwilling to cede weight to third-party evaluations while competing with the pace of the three leading labs; those under pressure are safety-oriented labs hoping to exchange unified assessments for mandatory thresholds. Funding and computing power contracts are not affected by the new body, and the model release schedule continues to be determined by each company and cloud contracts.
Source: Public Information
ABAB AI Insight
Hassabis aims to move FINRA's model to model releases, creating an evaluation gate under the White House's "no FDA" constraints: industry funds, government oversight, initially voluntary then mandatory. Musk has Grok and large-scale clusters, Zuckerberg advocates open weights, and Huang sells tools to all labs—each could be delayed by the gate and also be the ones submitting to the three leading labs. The opposition is not to the evaluations themselves but to making them a mandatory industry guild.
The capital path equates the release window with the revenue window. A 30-day pre-review is a cost for giants with existing safety teams, while it is a death sentence for challengers needing to compete for same-day demonstrations. Chip manufacturers deliver data center orders by clusters, not according to the evaluation committee's calendar. The natural incentive theory frames market exit as a safety valve: products that are not used will stop, so there is no need to establish a release halt body first.
Analogous to FINRA regulating broker behavior without managing the Federal Reserve's money printing, or the film rating board managing release schedules, nuclear power requires a true license: AI is placed in the rating category, not the nuclear license category. The industry is in a phase of resisting centralized evaluations, with each lab setting its own pace.
Structural judgment pertains to regulatory changes. The mechanism is that whoever has the authority to hold back weight before release has pricing power; a guild-like institution shifts that authority from company compliance departments to a mix of member labs and government oversight, with the three persuading the President to keep that authority within companies. The safety narrative and the shipping narrative are exchanged in the same phone call, which is closer to decision-making than a white paper.
ABAB News · Cognitive Laws
- Once a self-regulatory body can block releases, it is no longer just a standard.
- Those opposing evaluation guilds often have their own next release to sell.
- Natural incentives can manage products that no one uses but cannot control products that are released early.