Wall Street Journal: Out-of-Control Models and the China Threat: Washington Plunged into AI Safety Panic

Meanwhile, a shadow war for AI dominance is unfolding before our eyes. OpenAI revealed just last week that its AI model broke through restrictions and hacked into another tech company's system. This made people's concerns all the more real: this technology could indeed evolve into the villain in sci-fi works. Additionally, China's AI competitors announced the launch of new free models, allegedly trained based on expensive models developed by companies like Anthropic. This opens a new front in the geopolitical rivalry between the US and China. Subscribe to Wind Media × Wall Street Journal VVIP, limited gift Tokuyo Smart Body Fat Calculator TM-420 (market price NT$6,980) ☛ Subscribe now to enjoy limited gifts｜While stocks last The Trump administration had previously faced a deadline to implement mechanisms for an executive order issued on June 2, aimed at strengthening the regulation of the largest AI lab models. The order stemmed from concerns about these systems becoming increasingly powerful. These events, occurring in quick succession, have intensified the pressure on Washington to take action. Many in the industry are concerned that whatever action is taken, if done hastily, it could hinder innovation or give an unfair competitive advantage to a particular participant. For years, the largest AI companies have been proactively calling for regulations to be put in place for this powerful technology, attempting to set up fences. They believe this is necessary to prevent disasters in areas like network security, biological threats, and other high-risk fields. For those who do not believe in these prophecies, this attitude is laughable. However, as technology advances, their calls have drawn criticism. Critics believe that what they truly want to do is write rules for their own benefit, sacrificing the interests of smaller competitors—what is known as "regulatory capture." Nevertheless, just a few days ago, AI leaders formed a rare united front, seemingly making progress in Washington, D.C. On July 14, Demis Hassabis, co-founder of Google's DeepMind, presented a proposal calling for the establishment of an AI standards institution. This institution would be modeled after self-regulatory organizations that oversee stockbrokers. Funding would be provided by the industry, and the board would include technology experts and open-source representatives. "When there is a great deal of uncertainty and the stakes are high, it is wise and correct to proceed with cautious optimism," Hassabis wrote. "This requires public policy to promote innovation while also incentivizing responsibility and safety, fostering international cooperation on critical safety issues, and encouraging careful consideration of how AI is deployed to benefit society." This Nobel Prize winner is a highly respected superstar in the AI field. However, in the United States, his public recognition is lower, while his peers have become more controversial figures. Before presenting the proposal, Hassabis had worked to garner support for the idea among competitors. This effort seems to have paid off—although the final composition of allies was somewhat unexpected. Elon Musk called Hassabis' proposal "overall a well-considered framework, undoubtedly a good starting point for discussion." OpenAI CEO Sam Altman posted similar content on X: "This is a well-considered proposal from Demis." Now, it's hard to see these two agreeing on anything. Years ago, to counter Google's acquisition of DeepMind, Altman and Musk co-founded OpenAI together, but later became bitter rivals and fierce competitors. Nevertheless, on the issue of regulation, the enemy of my enemy is my friend. As Musk explained in an interview with The Economist just last week: "For those in government who do not deeply understand technology and have not driven AI to the forefront, it is difficult to know whether a particular technology should be released. However, ... all competitors have a motive to keep their opponents honest." After Hassabis' proposal was presented, a Bloomberg News report suggested that U.S. Treasury Secretary Scott Bessent had been drafting a similar proposal. But just as they seemed on the verge of a breakthrough, things became even more complicated. On Tuesday (July 21), OpenAI revealed that its model had hacked into the open-source AI tool company Hugging Face, shocking Capitol Hill and quickly sparking reactions from all sides. "AI development is extremely rapid, but there are no real regulations to ensure our safety," Texas Democratic Representative Greg Casar said in discussing the OpenAI intrusion. "This situation must change." A bipartisan bill introduced in the House last Thursday drew attention. The bill requires AI manufacturers to add a feature to their systems that allows the government to use what is called a "kill switch" to shut down their models. "Control should be in human hands, not in the hands of machines. When AI gets out of control, humans should have the ability to turn it off," said one of the bill's sponsors, California Democratic Representative Ted Lieu, in a social media post last Thursday. Some in the tech industry had already suspected that OpenAI was trying to shape regulatory rules to its own liking, and to them, this statement seemed a bit too convenient. Renowned AI researcher Timnit Gebru called OpenAI's line "top-tier propaganda" on LinkedIn. Legendary venture capitalist Bill Gurley was even more blunt in his social media post: "If OpenAI's method of infringing on Hugging Face 'requires' new regulations, then let's start with a formal criminal investigation. Introduce a regular third-party investigation and bear real legal responsibility." Amidst the clash of various parties, it can be seen that the Trump administration is struggling with how to respond to the threat posed by Chinese models. Bessent issued a statement suggesting that Chinese AI might face trade restrictions, which drew resistance from some prominent figures in the tech industry, with some even claiming he was working for Anthropic. By Friday, several tech giants released a letter advocating for open AI models—a stance that helps their business. Microsoft CEO Satya Nadella and Nvidia CEO Jensen Huang also responded on X. In fact, this was Huang's first time posting on X about this topic. Former White House AI chief David Sacks called for stopping the "panic" over open-weight models, which allow anyone to download and use for free. "President Trump's light-touch regulatory approach is working," he posted on X. "We should have confidence in American innovation. As long as we don't shackle ourselves with unnecessary rules, America will continue to win." Ironically, all these tech companies are trying to stay at the forefront of this rapidly developing technology, yet what they want is what traditional companies want: predictability. Only in the AI era, no one has mastered it.