According to the Financial Times (FT) and Reuters on the 21st (local time), President Trump told reporters at the Oval Office that he had postponed the signing ceremony of an executive order on AI scheduled in the afternoon. "I didn't like certain parts," he said. "We are leading everyone (in AI), including China. We don't want anything to block that initiative."
This executive order calls for major AI companies such as OpenAI, Google, and Anslogic to voluntarily submit their models to the government's national security and cyber risk checks. Specifically, it is designed as a voluntary cooperation system that relies on the goodwill of companies, not coercion, but 90 days before the state-of-the-art AI models are released to the public.
President Trump did not specify which part he did not like. However, he stressed that AI is "bringing enormous profits and creating many jobs," adding, "(This) executive order could be a stumbling block."
Big Tech CEOs were also scheduled to head to Washington to attend the signing ceremony, but it was canceled due to the sudden delay in signing.
The postponement of the signing was decided after weeks of internal strife within the administration over the scope of AI regulations. While some aides have argued that key models should be put under U.S. government control, some members of Trump's "MAGA" have warned that measures to restrict AI growth are harmful to the U.S. economy.
White House National Economic Council Chairman Kevin Hassett previously mentioned that the state-of-the-art models, like drugs, should be approved only after they are proven to be safe. However, the remarks were met with strong opposition from AI founders and investors, including those close to the Trump administration. This is because such regulations hinder U.S. innovation. In the end, the executive order ended up preparing a "spontaneous benchmarking cooperation framework" that is far from the mandatory approval requirements.
What ignited the regulatory debate was the Mythos model of Anthropic. Key U.S. officials, including U.S. Treasury Secretary Scott Bessant, are said to have been astonished at the vulnerability revealed in banking systems after examining the model in advance, which is highly capable of detecting cybersecurity vulnerabilities. Anthropic has restricted Mitos disclosure to a small number of trusted institutions, such as companies and some banks, before hackers access the model, allowing them to check and supplement cyber problems in advance.
Meanwhile, the decision to postpone the signing of the executive order came amid a series of polls showing that U.S. voters are concerned about the impact of AI and support intense regulations and safeguards, the FT noted.
In a survey conducted this month, 82% of Americans reportedly supported the White House's safety verification of advanced models. Public concerns about the safety of strong AI models, coupled with concerns about job losses and large-scale data center construction, are further complicating the administration's political calculation, which has been actively pro-AI until recently.
Bang Sung Hoon (bang@edaily.co.kr )
"It is a situation where the demand for security and the conflicting aspect of competition with China are colliding.
Concerns within the U.S. continue, but there are doubts about whether new industries can truly emerge if it falls behind in competition with China, which raises worries about hastily establishing regulatory policies.
Since this is an executive order rather than legislation, one might speculate that revision or repeal could be possible.
I hope the policies will shift to be as favorable as possible toward the AI sector."