AI needs a steering wheel more than it needs brakes

In a rare display of unity, leaders from Silicon Valley's artificial intelligence sector have collectively called for pacing of the frontier AI development. This brings the debate over how best to manage the rapid evolution of AI. The question is not whether to develop or govern AI, but how to do so effectively.

Dario Amodei, CEO of Anthropic, wrote that "we must pace the frontier (of AI)," a call that drew the unusual support of his rivals — OpenAI CEO Sam Altman and xAI founder Elon Musk. They argue that the iteration pace of frontier super-large models and autonomous AI agents must be controlled to buy time for risk governance.

This rare alignment among the three giants further highlights the deep divisions within the United States on the issue. A recent opinion poll showed that 61 percent of US respondents surveyed oppose building data centers to serve AI. Critics have targeted the White House's relatively lenient regulatory approach, suggesting that a more cautious stance is needed to address the potential risks associated with AI's rapid development.

At the All-in Summit in Los Angeles, NVIDIA CEO Jensen Huang took a call from US President Donald Trump, putting him on speakerphone in front of the audience. Trump dismissed AI alarmism as a "hoax" and said that "robots will not be taking over the world".

Meanwhile, US Vice-President JD Vance described calls for stronger regulation as a "Trojan horse", implying that AI giants want to use rules as walls to block their competitors.

Concerns about AI development outpacing human control are not entirely unfounded.

The recent OpenAI Hugging Face incident, where an AI agent acted on its own initiative and attacked an unauthorized target, shows that frontier large models are gradually gaining self-improvement capabilities.

The crisis, resolved with the help of a Chinese company, highlights the potential systemic risks posed by AI. Unlike traditional software, where errors are fixed vulnerabilities, the behavior of large-model agents can be unpredictable and potentially lead to catastrophic damage.

While calls for regulatory guardrails reflect a cautious approach, there is skepticism about the motives behind the industry's "slowdown" narrative. It is seen not as a fair choice that treats all participants equally.

Leading US AI companies currently control vast computing resources, possess deep capital reserves and have the most complete software ecosystem.

If the AI industry hits the brakes, these leaders would lock in their advantage while others lose the chance to close the gap. Barriers on the track would harden, potentially creating an oligopolistic structure.

Notably, neither the pro-pacing nor the anti-pacing narrative conceals its geopolitical position. Amodei emphasizes maintaining the US and its allies' leadership, while opponents contend that slowing down the pace of development would benefit US' competitors.

On this point, both camps take the same stand. A US media outlet has analyzed that both advocating and opposing the pacing of frontier AI are intended to enhance the bargaining leverage against competitors.

There is global consensus that the risks presented by AI's rapid development are different from those posed by other new technologies. Whatever the motives of the Silicon Valley giants, it is hoped that their calls for a slowdown will advance the discussion on global AI governance and promote cooperation among nations.

Earlier this month, a European theoretical physicist released a video titled "Someone Paid Me to Tell You AI Will Kill Us". The scientist contended that both sides have financial backers: some pay to dramatize existential risk, while others pay to sell techno-optimism.

The scripts and arguments are prepared in advance, all driven by vested interests.

As a Chinese Foreign Ministry spokesperson noted, "The faster AI technology develops by leaps and bounds, the more firmly its direction must be correctly anchored to goodness and to benefiting humanity."

For a technology advancing at high speed, a steering wheel is more important than brakes — and probably easier to operate as well.

According to Stanford University's AI Index Report 2026, 84 percent of Chinese respondents feel excited about AI, compared with only 38 percent in the US. Last year's report showed that 83 percent of Chinese people believed AI products would do more good than harm, while only 39 percent of US respondents held that view.

Obviously, AI enjoys greater social acceptance in China, possibly due to the Chinese government's consistent advocacy of the principles of "people-centered, intelligence for good, equality and inclusiveness, safety and controllability, and collaborative governance," and its practical efforts to make AI a new driving force for common prosperity and security.

How can we get along with thinking machines? And is it possible to ensure security when algorithms are part of decision-making? How can we tackle ethical challenges posed by technology through adaptive governance? And how can we realize AI for all when the divide keeps widening?

President Xi Jinping addressed these issues at the 2026 World Artificial Intelligence Conference, calling for a collaborative effort to build a just and equitable system for global AI governance.

The author is chief researcher at the China Watch Institute, China Daily.