The Truth Behind the AI Doomsday Debate: A Power Play by Five Men

The debate over AI doom looks like a discussion of humanity’s survival and technology ethics. In reality, it is a contest of interests over leadership in the development race and fundraising. Both those calling for a slowdown and those demanding acceleration have their own clear positions. Reporting by Shangyinsha of Huxiu describes this confrontation as a play performed by five men. It is necessary to sort out the commercial and political calculations behind the banner of safety.

Behind the Phone Exchange in Los Angeles

On September 14 local time, the All-In Summit was held in Los Angeles. While NVIDIA’s Jensen Huang was on stage, his phone rang. The caller was the U.S. president. Huang switched the call to speakerphone and shared it with the audience.

President Trump said robots will not take over the world. He declared that artificial intelligence will absolutely never come to dominate the world. He indicated that all recent talk about loss of control is a scam. He also said data centers bring wealth to local communities. He pointed out that if the United States slows down out of concern, China would be the happiest.

“Robots will never take over the world, and artificial intelligence will never dominate the world. That will absolutely never happen.”

Huang responded that he would not let a slowdown happen. Applause immediately spread through the venue. This exchange was a head-on rebuttal to AI doomsday theory. Behind it are the growing calls for deceleration and the political calendar.

The Full Scope of Amodei’s Slowdown Proposal

The dispute began with Anthropic’s move on September 12. CEO Dario Amodei published a lengthy essay. Its title conveyed that the frontier’s advance must be slowed. He warned that at the current pace, danger could emerge within 6 to 12 months.

“At the current pace, within 6 to 12 months, swarms of autonomous AI agents could gain control through botnets.”

He estimated that losses could reach hundreds of billions of dollars. He then called for slowing the advancement of frontier models. He argued that safety research needs to catch up. There are three specific measures. Embedded third-party evaluations, industry-wide common safety standards, and global-scale coordination.

That same evening, OpenAI’s Sam Altman concurred. He said the pace of frontier development needs to be managed. Then xAI’s Elon Musk also reposted it to show agreement. Three men who are direct competitors or long-time rivals lined up together. This alignment drew attention.

The Contradiction Between the Safety Banner

and Rolled-Back Safeguards

Anthropic and OpenAI have championed safety as their banner. At the same time, they have a history of rolling back safeguards. The gap between words and actions has fueled the current doubts.

OpenAI at its founding set as its mission safe artificial general intelligence that benefits humanity. It also explicitly stated it would not be constrained by the need for financial returns. Yet in corporate documents at the end of 2025, that crucial language had been removed. In January 2024, it was reported to have lifted the ban on military and warfare uses of ChatGPT. Amid growing AI demand from the U.S. military, it effectively softened its previous stance. It also abolished its transparency principles, making external verification difficult. In May 2024, five safety staff members resigned. One of those who left publicly warned about a flashy, product-heavy focus.

Anthropic also placed safety at its core when founded. Amodei had served as vice president of research at OpenAI. Driven by doubts about its values, he left in 2021 with a group of core staff. Adopting a company name meaning humanity, he created Claude. He announced a method of steering outputs with constitutional principles. In September 2023, it introduced its Responsible Scaling Policy. It was a mechanism to stop training if catastrophic-risk thresholds were reached. But in version 3.0 of that policy this February, that promise was withdrawn. The condition for halting training without prior mitigations disappeared. It was replaced with a milder framework centered on roadmaps, risk reporting, and external reviews.

The Commercial Motives Behind the Three

Slowdown Advocates

The slowdown camp’s arguments are each said to have a commercial background. Anthropic is at a critical moment ahead of a listing. Just days before its report in June this year, it had quietly filed for an initial public offering. Its valuation was reported to be near $1 trillion. Its safety banner since founding is central to differentiating itself from rivals. In a listing phase, the appeal of the story is put to the test. Strengthening the safety narrative directly translates into a higher valuation.

Altman once criticized Anthropic’s theatrics. He quipped it was building bombs while selling bomb shelters. Even so, this time he joined the slowdown camp. Behind that lies a performance gap. OpenAI posted a net loss of $38.5 billion in 2025. Anthropic’s full-year net loss was about $4.2 billion. In the first half of 2026, Anthropic’s revenue growth far outpaced OpenAI’s. In the second quarter, its quarterly revenue exceeded its rival’s for the first time. Anthropic turned profitable, while OpenAI remained in the red with slowing growth. Now is not a good time for OpenAI to go public. Prioritizing safety is a convenient explanation for a delay. His shift from acceleration to deceleration lies in that context.

Musk has long warned about the dangers of AI. For xAI, in a chasing position, a slowdown works in its favor. The proposal for mutual testing is seen as intended to hold back frontrunners. The alignment of bitter rivals is not a meeting of ethics but of interests.

Unpacking the Positions and Political

Background of the Two Accelerationists

The acceleration camp also has clear positions. Huang depends on the exponential growth of AI infrastructure. NVIDIA is a supplier of computing infrastructure for AI. A slowdown would be a direct blow. Hence his opposition to slowing down has been consistent.

President Trump’s remarks have two backgrounds. One is maintaining an edge in competition with China. He framed a slowdown as benefiting China. The other is a domestic pocketbook issue. There is discontent that data centers are pushing up electricity bills. Ahead of the midterm elections, he needed to dismiss that criticism as misinformation. His claim that data centers bring wealth is part of that effort.

The Structural Problem of Unverifiable

Doomsday Risks

It is true that AI development comes with safety issues. But today’s AI is still in the early stages of a long journey. Whether the danger is human-extinction-level remains abstract and questionable. Many claims can be neither disproven nor verified. No concrete scenario for an imminent crisis has been presented. While emphasizing crisis, solutions are scarce. Rather, they have advanced development while removing safeguards. Doomsday risk is close to a tool for fighting over market voice. Anthropic’s report in June this year has the same structure. It warned that AI has begun to build itself. It said recursive self-improvement would advance within two years. It called for delaying or halting frontier models. Meanwhile, business expansion and IPO preparations proceeded in parallel. The consistency between words and actions is being questioned.

Editorial Opinion

In the short term, we expect the development race for frontier models will not slow, while evaluation and safety standards are developed in parallel. Embedded third-party evaluations and debate over common standards could be built into procurement and partnership conditions. We assess the data-center electricity-price issue will become politicized ahead of the midterm elections. Companies will continue to use the safety narrative to manage fundraising and IPO timing.

In the long term, we expect the strategy of using safety as a differentiator to take hold, with review systems valued over transparency. There is also a view that mutual-testing and international-coordination frameworks will become entry barriers favoring large players. If unverifiable doomsday risks keep being invoked, we assess the basis for policy decisions will become blurred. Users and developers will need to distinguish concrete-harm measures from abstract concerns.

The question from the editors is at whose expense slowdown claims will be implemented. We see the costs of halts or delays potentially falling disproportionately on startups and research institutions. We also assess it remains unclear who will hold the power to define safety and evaluation authority. The question is whether verifiable risks and countermeasures alone can be debated, setting aside profit calculations.

References

• “ 五个男人的一台“AI末日”戏 “, by 商隐社 — 虎嗅网, 2026-09-17T13:28:42.000Z (ARR)
• Source URL: https://www.huxiu.com/article/4892111.html?f=rss