The debate, which seems to center on the destiny of humanity and the ethics of technology, is, in reality, disconnected from both.

If a threat could truly wipe out humanity within a few decades, mere debate would not suffice to address it. Conversely, if an issue can be settled through debate, it hardly qualifies as an apocalyptic scenario.

Author | Haoran This article is an original work by Shangyin Club. For reprints, please contact the editorial team.

The Hidden Interests Behind the Concerns

On September 14 (local time), during the All-In Summit in Los Angeles, Jensen Huang was engaged in a live panel discussion on stage when his phone rang. It was the President of the United States calling.

He switched the call to speakerphone, allowing the entire audience to listen in.

Trump stated from the other end:

"Robots won't dominate the world, and artificial intelligence won't control the globe. That will never happen." He dismissed recent claims about AI going rogue as "all a hoax."

He also mentioned that data centers could generate wealth for local communities and warned that if the U.S. slowed down due to concerns about risks, China would be the most pleased.

Huang replied, "You're right. We won't let that (AI slowdown) happen." The audience immediately burst into applause.

The reason behind Trump's remarks is the recent "AI apocalypse theory" that has been making headlines.

This debate was sparked by Dario Amodei, CEO of Anthropic, who published a lengthy article titled "We Must Slow Down the Frontier" on September 12. He cautioned that "at the current pace, within 6 to 12 months, a group of autonomous AI agents could gain the ability to take over the entire internet through botnets, causing hundreds of billions of dollars in damage."

Therefore, he called for a slowdown in the upgrade of the most advanced models to allow safety research to catch up and proposed a three-step plan—introducing embedded third-party evaluations, establishing industry-wide safety standards, and promoting global coordination.

That evening, Sam Altman of OpenAI responded:

"I agree with Dario. We need to control the pace of frontier development."

Immediately after, Elon Musk of xAI retweeted with the comment:

"Dario is right."

What is the connection between these three individuals? Amodei and Altman are direct competitors; Musk and Altman have been publicly estranged for years, with Musk even suing OpenAI and the two regularly clashing on social media.

Yet, at this moment, adversaries and rivals have united on the same side, calling for a slowdown in AI.

They may genuinely be concerned about the threats posed by AI, but historically, such concerns have been rare. As the current "twin stars" of the AI world, Anthropic and OpenAI have consistently emphasized safety while simultaneously dismantling safeguards to accelerate their progress.

When OpenAI was first established, its mission statement explicitly aimed to build artificial general intelligence that is "safe and beneficial to humanity, unrestricted by the need for financial returns." However, by the end of 2025, key phrases like "safe" and "unrestricted by profit motives" had been removed from the company's documents.

In fact, OpenAI had already prioritized profits over safety before this:

In January 2024, media reports revealed that OpenAI had quietly lifted the ban on "military and warfare applications" for ChatGPT. As the U.S. military's demand for AI technology grew stronger, OpenAI subtly weakened its previous stance against military use;

At the same time, OpenAI also abandoned its long-held principle of transparency, leaving external observers without a factual basis to assess OpenAI's governance structure, conflicts of interest, and safety decision-making mechanisms;

In May 2024, five employees responsible for safety-related work resigned. Among them, Jan Leike warned the public that the company placed too much emphasis on "flashy products" while neglecting AI safety.

Similarly, Anthropic initially placed safety at the core of its mission.

Dario Amodei, formerly OpenAI's Vice President of Research, left OpenAI in 2021 along with six other core employees due to doubts about OpenAI's values and concerns over safety. They founded Anthropic (meaning "human") and launched the star product Claude.

From the beginning, Anthropic positioned itself as the world's most serious "AI safety" frontier company and released "Constitutional AI" to train AI—by creating a set of explicit value principles, allowing AI to use these principles to judge and correct its own outputs.

Additionally, in September 2023, Anthropic introduced the "Responsible Scaling Policy," a governance framework to assess the catastrophic risks models might pose. If certain standards were met, the company would voluntarily pause training until safety was ensured.

However, in the 3.0 version of the "Responsible Scaling Policy" released by Anthropic in February of this year, the previous commitment to "not train more powerful models without ensuring risk mitigation measures are in place beforehand" was removed. Instead, a softer transparency mechanism centered on roadmaps, risk reports, and external reviews was introduced.

Admittedly, AI development does bring a series of safety issues, but currently, AI has only taken the first step in a long journey. Are the risks truly at an "apocalyptic" level that could destroy humanity?

This is highly abstract and quite doubtful—it is "unfalsifiable" and cannot be verified. No tech company has provided a credible, specific scenario proving that such an apocalyptic crisis is imminent. They merely emphasize the crisis without proposing any substantial solutions, all while continuously dismantling safety safeguards to rush ahead.

This inevitably leads one to believe that behind these concerns for humanity's survival lie considerations of interest.

A Drama Staged by Five Individuals

In June of this year, Anthropic warned in its report "When AI Begins to Self-Construct" that AI systems could achieve recursive self-improvement within two years—autonomously designing, training, and improving successor models, with gradually diminishing human intervention in model training. Therefore, the report called on major global AI labs to slow down or even pause frontier model development.

However, just days before this report was released, Anthropic had secretly filed for an IPO with a valuation nearing one trillion dollars.

Now, with Dario Amodei once again calling for a slowdown in AI, it coincides with Anthropic's critical period of going public.

As is well known, the IPO window period requires a higher level of narrative capability from companies. Since its inception, Anthropic has emphasized "safety," which is also its core selling point distinguishing it from competitors. By continuously reinforcing the "safety narrative," it can further inflate its valuation premium.

Sam Altman has sharply criticized Anthropic's self-marketing tactics: "We built a bomb, are about to drop it on your head, and then sell you a $100 million bomb shelter—but only to the people we choose. This is obviously incredible marketing."

But criticism aside, this time Altman has also aligned with Amodei, not out of concern for humanity's fate, but because he too needs the "safety narrative" at present.

OpenAI reported a net loss of $38.5 billion in 2025, while Anthropic's annual net loss was approximately $4.2 billion. By the first half of 2026, Anthropic's revenue growth had far surpassed OpenAI's, overtaking it in quarterly revenue scale for the first time in the second quarter while achieving profitability. In contrast, OpenAI not only continued to incur losses but also saw its growth slow down.

So for OpenAI, now is not a good time to go public. A decision to delay the IPO can be framed in two ways: "We are delaying going public for the sake of human safety" or "We are waiting for a better valuation window." Altman clearly chose the former as his public statement, swiftly switching from an "accelerationist" to a "decelerationist."

Thus, when Fortune recently interviewed Altman and asked whether he felt pressured to rush the IPO, his response was decisive:

"We won't rush to go public. Considering everything happening in the safety domain, now would be an unwise time to go public. We feel no pressure to do so."

Compared to Altman, Musk has consistently emphasized the dangers of AI. As early as 2014, he believed that "AI could be more dangerous than nuclear weapons" and even predicted that the probability of AI leading to human extinction was around 20%.

Later, he continued to warn about the dangers of AI but also founded xAI, with the core logic being: Since AI is inevitable, rather than worrying from the sidelines, it's better to participate and ensure AI develops in a direction he considers safe.

However, xAI is a pursuer in the frontier model competition, and calling for a "slowdown" benefits pursuers. Just as he expressed agreement with deceleration, xAI's Grok 4.8 pre-training had just concluded, and reinforcement learning training was about to begin.

Meanwhile, at the All-In Summit on September 15, Musk proposed a scheme for regulating AI: Leading AI companies (including xAI) would test each other's models before official release, with competitors responsible for identifying and raising safety concerns.

This also reflects Musk's calculations: He wants to slow down the leaders through mutual supervision, which benefits the pursuers, while also participating in setting industry rules. Industry self-regulation can also avoid government intervention in AI regulation.

The three figures in the "decelerationist" camp share a common view but each has their own agenda.

Jensen Huang, as the "shovel seller" in the AI industry, is a staunch "accelerationist." NVIDIA's earnings guidance relies on "exponential growth in AI infrastructure," and deceleration rhetoric directly threatens this narrative. NVIDIA would have a different answer on the deceleration issue.

So he vigorously argues that the predictions of AI pessimists have not materialized on schedule: Radiologists would be replaced by AI within a few years; 90% of code would be generated by AI in the short term; and large numbers of junior positions would quickly disappear.

Another heavyweight in the "accelerationist" camp is Trump, who repeatedly mentions one judgment in all relevant statements: The U.S. is currently "about a year ahead of China" in AI, but this advantage could be lost at any moment.

He has explicitly said: "I'm not worried at all (about AI leading to human extinction)." "What I'm worried about is that if we can't win the AI race, we'll be in a very disadvantageous position.

This anxiety directly drives his actions. In May 2026, he was prepared to sign an executive order requiring frontier AI models to undergo a 90-day government safety review. However, David Sacks, the former White House AI affairs chief, made an emergency call on the day of signing, warning that the review mechanism would "severely slow down the innovation speed of the U.S. AI industry and even cause the U.S. to lose to China in the AI competition." Subsequently, tech executives like Musk and Zuckerberg also called to apply pressure, and Trump ultimately canceled the signing at the last moment.

Additionally, Trump's strong opposition to the "AI deceleration theory" is also aimed at the upcoming U.S. midterm elections. The Trump administration has generally adopted a hands-off approach to AI, but American public dissatisfaction with AI construction has grown, especially regarding data centers, which are blamed for driving up electricity prices.

In March of this year, an NBC News poll showed that AI's approval rating was even lower than that of the U.S. Immigration and Customs Enforcement (ICE). A May Gallup survey also found that seven in ten Americans opposed building AI data centers in their local areas.

For this reason, while emphasizing that claims about AI posing fatal risks and calls for government regulation are "absurd conspiracy theories," Trump also mentioned that "data centers are the oil of the next 20 to 25 years, far more valuable than the internet and much more important.

He aims to define a livelihood issue unfavorable to his party as a "hoax" before the midterm elections.

Ultimately, the debate that seems to center on humanity's destiny and the ethics of technology is, in reality, disconnected from both.

If a threat could truly wipe out humanity within a few decades, mere debate would not suffice to address it. Conversely, if an issue can be settled through debate, it hardly qualifies as an apocalyptic scenario.

Welcome to exchange and discuss with the author~