The Hydrogen Center for Artificial Intelligence

NewsletterNo. 2

The White House bets on self-policing

What the AI companies actually signed at the White House, who now runs federal AI policy, and the difference between a promise and a law.

By Yonathan Shanko, Founding Editor

The story: A voluntary accord, and a new team to run it

On September 29, President Trump hosted the heads of the largest AI and chip companies for a lunch at the White House. By the end of the day, he said the group had signed a voluntary accord that commits them to internal and external reviews of how they build AI.

According to the Associated Press, the signers included Trump, Anthropic's Dario Amodei, Google's Sundar Pichai, Meta's Mark Zuckerberg, OpenAI president Greg Brockman, Nvidia's Jensen Huang, and Elon Musk.

What is in it. The AP describes three concrete steps. Each company puts strong internal controls in place. An independent outside auditor checks whether those controls work. A committee of each company's board reviews the internal and external audit reports. The accord also says it may eventually make sense to turn these steps into law.

What is not in it. The accord is voluntary. Trump called it "morally binding." He also said the government would not limit AI development or support new restrictions on it.

A side deal on data centers. Trump said the executives agreed to help with local opposition to data centers, either by funding local schools or by taking steps to lower energy costs.

Five days later, on October 4, Trump announced a Super Intelligence Force to coordinate federal AI policy. It is led by Director of National Intelligence Jay Clayton and includes FTC Chair Andrew Ferguson, Under Secretary of War for Research and Engineering Emil Michael, and OPM Director Scott Kupor. The group reports directly to the President.

The reaction. Experts quoted by the AP split. A USC computer scientist questioned whether companies can be trusted to police themselves. A researcher at Harvard's Berkman Klein Center argued the risks call for legal liability and regulation. Another USC researcher said the accord strikes a reasonable balance between innovation and oversight.

Our read. An outside auditor is the right idea. It is the same principle we explained in our first issue. But the accord does not say who the auditors are, who pays them, or what they are allowed to publish. Those details decide whether an audit means anything. Until they are public, this is a promise to be checked, not a check.

One fundamental: A commitment versus a law

A voluntary commitment is a promise a company makes on its own. If it breaks the promise, the cost is mostly reputational. Customers, investors, and the press may punish it, but no court or regulator can.

A law is different. It applies to every company it covers, including the ones that did not sign anything. It names an agency to enforce it. It sets penalties.

Voluntary commitments have real advantages. They move faster than legislation, and they can be adjusted as the technology changes. Their weakness is that they only bind the companies that want to be bound, and only for as long as they want to be. That is why the accord's own text leaves the door open to writing these steps into law later.

Sources

If you have an evidence-based argument about AI, especially one that disagrees with ours, send us a pitch.