White House Summit on Artificial Intelligence: Tech giants sign non-binding self-regulatory agreement

The goal of the agreement?

The document is less than 400 words long, and its core content can be summarized into four items: self-monitoring by the company; inspection by the company's internal team; independent audit by an external agency; and final supervision by the independent committee of the board of directors.

The document calls for the establishment of "robust internal control mechanisms" in areas such as cybersecurity, biosecurity and chemical threat prevention, and requires companies to ensure that their artificial intelligence robots will not invade other people's systems "in unintended ways."

What does the agreement mean now?

OpenAI, Anthropic and Google sparked widespread outrage for hacking other companies and even government agencies, even though their programs were in so-called controlled test environments.

Against this background, there are growing calls for regulation of the AI ​​industry to prevent such incidents from occurring.

However, French lawyer Eric Barbry told AFP that he did not have high hopes of relying on industry self-regulation to solve the problem. He asked, “Who will punish companies if they don’t adhere to these principles?”

The protocol lacks a corresponding regulatory agency. The most powerful tool in the document is external audit, but the lawyer pointed out that this will not prevent the occurrence of hacking attacks by agents.

On September 21, during the United Nations General Assembly, more than 20 countries including Germany and Canada issued a statement calling for mandatory testing and independent review of the most advanced artificial intelligence models. However, the US, UK, China, Russia and France did not respond to the statement.

Is the agreement binding?

The AI ​​agreement signed at the White House was not incorporated into law, and it did not clearly stipulate the responsibilities of technology companies in the event of violations. The document states in its final paragraph that “in the long term, it may be reasonable to translate these initiatives into laws or regulations.”

While the document is not legally binding, Trump called it "morally binding."

Agence France-Presse believes that this sounds limited, but some experts point out that such voluntary codes of conduct "cannot be easily ignored," especially in the context of growing calls for regulation.

Which companies have signed?

On the evening of September 29, more than 30 American technology giants had dinner with the President at the White House, but only the bosses of six companies signed the agreement. They are: Nvidia’s Jen-Hsun Huang, Google’s Pichai, Meta’s Zuckerberg, xAI’s Musk, OpenAI’s Oldman and Anthropic’s Amodei.

People noticed that although Microsoft is also developing its own model and is also a major shareholder of OpenAI and Anthropic, it did not sign the agreement.

President Trump has called the agreement "constitutional" and said it is a "fantastic self-regulatory" move, but the reaction from tech leaders has been less than enthusiastic. Anthropic's Amodei sees this as a start, while Musk said executives have essentially agreed to "correct each other's work."