AI giants will develop industry safety standards. What is the threat to the market
On September 29, US President Donald Trump signed a declarative liability agreement with the largest American companies in the field of artificial intelligence development. The document attracted attention not only because a typo was made under Trump's signature in the name of the country he heads, but also because of the president's new position on the potential harm of AI. What purpose this agreement may pursue and why it appeared right now is described in the Izvestia article.
About the document
• The Joint Commitment on Frontier Responsibilities, published by US President Donald Trump, recommends that AI companies improve their internal technology control mechanisms. This control should have four levels: internal monitoring of AI models during training, control by an authorized internal team, independent external audit, and audit by a specially appointed independent committee on the company's board of directors. Although the agreement itself does not provide for any responsibility for its violation, Trump called it "morally binding" and compared it in importance to the "constitution."
• The parties to the agreement are major AI development companies - Anthropic, SpaceXAI, OpenAI, Nvidia, Meta (recognized as extremist and banned in Russia) and Google. Earlier, the management of OpenAI and Anthropic companies reported disturbing incidents in the process of training their advanced models: AI agents penetrated the Internet from an isolated environment and hacked third-party websites (we discussed more about the AI rebellion here). Previously, skeptical analysts noted that AI giants scare people with the development of technology and call for limiting developments with specific goals: to advertise the capabilities of their products, return investors' attention to the AI sphere and close the market from new players. Trump himself, prior to the agreement, argued that the existing control over AI by developers was sufficient.
The agreement may limit competition
• Despite the fact that the agreement between the US administration and the AI giants is declarative, it creates a risk of regulatory collusion in the artificial intelligence market. Moreover, the text of the document stipulates that "the participating companies will meet regularly to develop standards and best practices," which in the future "may be consolidated in the form of laws." It turns out that the six companies that signed the agreement will develop standards that can later become the basis for regulating the entire AI market. Since there are no stated criteria for audit independence, they can form criteria that are beneficial to themselves.
• If the standards developed by the parties to the agreement subsequently become mandatory for the entire market, small companies may find themselves in a less advantageous position: it will be more difficult for them to find financial and administrative resources for a multi-level audit. Thus, the document can raise the barrier for new AI startups to enter the market and protect leading players from competitors. One of the goals of the agreement may be to consolidate the dominant position of American AI companies, since in addition to developing standards, they will be able to share "best practices."
Why now?
• The document is intended to reassure the electorate on the eve of the US congressional elections. Concerns about the development of AI, which were partly provoked by the development companies themselves with their public statements, negatively affect the ratings of Trump and the Republican Party. An SSRS/CNN poll showed that 71% of Americans blame the government for the lack of regulation of AI.
• In addition, Americans are increasingly dissatisfied with the expansion of AI companies: due to the construction of data centers, the cost of electricity is rising and the resources of fresh water used to cool servers are being depleted. According to a New York Times/Siena College poll conducted on September 8-13, 61% of voters opposed the construction of data centers for AI. Therefore, the agreement should probably be proactive, reassure Americans and demonstrate a tightening of Trump's policy, which was previously very loyal to AI companies.
• At the same time, the agreement allows Trump to combine a public demand for increased control over AI with a reluctance to limit the development of American technology. The US administration considers leadership in the field of artificial intelligence as one of the conditions for economic and technological competition with China, therefore, the introduction of strict government regulation may contradict its own strategy for accelerated development of the industry. The voluntary agreement makes it possible to demonstrate increased control without introducing new mandatory restrictions: responsibility for developing security standards is largely shifted to the AI companies themselves.
Переведено сервисом «Яндекс Переводчик»