Should AI Hit the Brakes: Divergent Positions and the Battle Over Commercial Motives Among Tech Giants

The debate over whether AI development should slow down has reached unprecedented intensity. In addition to several AI companies speaking out one after another, more and more people are taking sides, and the "AI doomerism" debate has extended from technology to the capital markets. The latest remarks came from Meta CEO Zuckerberg. On September 16, he publicly stated that using the vast majority of compute to serve people, rather than racing to pursue recursive self-improvement in AI, is one of the best ways to ensure the safe development of this technology.

The matter began to ferment when Anthropic CEO Dario Amodei published "We Must Pace Frontier AI" on September 12. In the article, he wrote that roughly since this summer, AI's ability to build the next generation of AI through "recursive self-improvement" has been continuously strengthening, and if left unconstrained, it could exceed humanity's ability to understand and control these systems. Dario worries that within 6-12 months, such a cluster could take over the entire internet through continuously running botnets, and even cause hundreds of billions of dollars in losses. He called on the industry to proactively slow the pace of improving frontier model capabilities.

This initiative unusually received recognition from competitors. Including OpenAI CEO Sam Altman, xAI founder Elon Musk, DeepMind founder Demis Hassabis and others all publicly expressed support.

But the voices of opponents are also very strong, among which U.S. President Trump and Nvidia CEO Jensen Huang are the most intense.

Trump believes AI safety concerns are a "hoax." On September 14 local time, Trump said in a phone call with Jensen Huang that robots will not take over everything, AI will not take over the whole world, and it is all a hoax. Jensen Huang expressed agreement.

The next day, while attending Salesforce Dreamforce in San Francisco, Jensen Huang once again spoke with a clear stance. He believes that innovation and safety are not an either-or choice, that the two can be achieved together, and that AI safety does not require new legislation. Therefore, we should move forward as quickly as possible, and if at a particular moment a company is deemed out of control or its products unsafe, then pause.

Jensen Huang said he envisions a future in which "every company, every enterprise... will become an AI company." He said companies will become more ambitious after having AI, and reiterated that those who do not accept this technology will be left behind. He also rebutted the claim that AI will destroy jobs, calling it "complete nonsense."

Soon after, Zuckerberg also joined this great debate. On the 16th, he said on social platforms, "There is a lot of debate about slowing down capability progress before alignment catches up. My view is that trust and alignment are rapidly becoming the most important capabilities distinguishing agents and models. Any lab that does not focus on alignment will fall behind." Zuckerberg believes that every lab has the responsibility and incentive to advance at the speed needed to train its models safely, and also has the ability to take action on its own to ensure this. People will not be willing to use agents that are not aligned with them and do not act as they require, so labs have a strong natural incentive to make their models more aligned.

Zuckerberg believes that using the vast majority of compute to serve people, rather than racing to pursue recursive self-improvement, is one of the best ways to ensure the safe development of this technology. In addition, independent evaluators and advisors can be introduced. "Overall, having a larger and more diverse ecosystem of evaluators will be very helpful."

Regarding Zuckerberg's remarks, Musk commented that it was "cool."

Saying brake on the lips, stepping on the gas underfoot

At the Dreamforce conference held on the 15th, Dario used a car metaphor to explain his view: when a competing automaker has a safety incident such as brake failure, this is the moment for all automakers to stop and examine their own practices. Dario said Anthropic has committed to introducing independent evaluators.

"It is very tempting to attack competitors and say these people are unsafe," he said. "But I think a more responsible response is: look at our own record. We may not have had this kind of major, high-profile incident, but I am sure we are not perfect."

Although Altman responded to Dario's initiative at the first opportunity, some differences have already emerged in his subsequent further statements.

After Dario's above remarks, Altman, who appeared in the Dreamforce fireside chat session, said that safety commitments should not be premised on "others being responsible," and there should be no conditions attached.

Altman said he believes companies and the industry have the ability to do this safely. "These models are becoming so good and so fast that we need to treat alignment, monitoring and safety with a new level of rigor. We must be willing to adjust the pace of capability development so that alignment, safety and monitoring always stay ahead of capabilities."

He said that if it reaches a point where this cannot be done, then slow down or stop. In all the ways the entire industry and society use new technology, some degree of incident is inevitable. What he really cares about is having a good culture of incident reporting and learning. He believes the hacking incident shows that companies need to prepare for the "coming wave of cyberattacks."

Well-known U.S. stock market "big short" Michael Burry directly pointed to the commercial motives behind this safety discussion. "Let us all take a moment to understand how self-serving it is for executives of OpenAI, Anthropic and other hyperscalers to talk about slowing down." On September 14, Michael Burry added a new perspective on social media.

He listed four judgments: First, large language models are not AI, nor will they become AGI, and there is no AI that needs to be slowed down. Second, competition is rapidly closing in, and slowing down benefits the incumbents. Third, an IPO requires hype and boasting, and "we are so powerful that we might become dangerous" is itself hype and boasting. Fourth, it provides cover for a slowdown in growth that is truly uncontrollable, because the IPO appears to be being delayed.

Before Michael Burry's analysis, Altman said in September that, out of concern for AI safety, OpenAI would not go public in 2026. "In fact, given everything that is happening on the safety front right now, going public now may not be a wise move, and we are not under pressure to do so."

On the other side, although AI doomerism has become rampant, multiple sources say Anthropic may still go public in 2026. It has chosen Nasdaq as the location for a potential IPO, and Anthropic believes that the transparency of a public company can improve safety. Anthropic has also finalized $517 billion in compute deals over the past 11 months, with compute capacity of 14.8 GW, and this compute will be brought online in the coming years.

According to Axios, another view holds that "pacing" may reduce Anthropic's compute spending, thereby improving its financial condition. Finally, companies still need capital. It can delay and raise more funds in the private market, but no frontier lab can stay outside the capital markets.

1; When the mathematical fortress is breached, can other disciplines usher in an AI breakthrough moment?

1; Paper topic inspiration: "Sustainable Development: Agricultural and Forestry Waste Treatment and Reuse" research direction | MDPI Sustainability

2; LAF From "Intelligent Teaching Assistant" to "Discipline Engine": How AI Is Reshaping the University Classroom

3; Paper topic inspiration: "Glioma" research direction | MDPI Brain Sciences

4; Soil moisture thresholds affect soil carbon dioxide emissions: A short-term soil carbon dioxide emission prediction method for climate-smart farmland based on machine learning | MDPI Sustainability

5; "Maize Sciences" | A review by the team of Researcher Fan Xingming of the Yunnan Academy of Agricultural Sciences on the breeding application research progress of the genetic basis of maize resistance to southern leaf blight and the molecular mechanisms of disease resistance

6; FDE The "takeover" of knowledge transmission and the "repositioning" of ability cultivation—the dual variation of education in the AI era

7; Recommended reprints from the journal Brain Sciences

8; Literature list: 2025 cover articles | MDPI Metals

1; The western cryosphere is accelerating its "degradation": glacier melting sounds the alarm for alpine safety

2; The 6th National Agricultural Science and Technology Development Strategy Think Tank Alliance Conference was held in Lanzhou

3; The 2026 Nobel Prizes are about to be announced, and authoritative experts will provide you with a comprehensive interpretation!

8; Without a paper, he still earned a doctorate and became the first person in the national medical field

Famous doctoral dissertations 163: Nobel laureate Dr. Peter Doherty's doctoral dissertation, University of Edinburgh, 1970

My elite university complex and my elite university experience

More >>