In the AI world’s three great enemies, for the first time in history they have unexpectedly stood on the same side—after a series of runaway incidents and successive doomsday warnings struck the industry, the heads of Anthropic, OpenAI, and SpaceX have rarely joined forces to call for slowing down AI development.
On Saturday, Anthropic CEO Dario Amodei published a blog post urging the global AI industry to voluntarily slow down, arguing that the pace at which current frontier models’ capabilities are improving has far outstripped humanity’s understanding and ability to control them. Within hours of the post going out, SpaceX CEO Elon Musk took to X to state, “Dario is right.”
OpenAI CEO Sam Altman immediately followed up, saying he agrees with Amodei’s assessment about “controlling the pacing of frontier progress,” and promising that OpenAI will open “employee-level” system access to independent third-party evaluation organizations.
Altman also told (Fortune) magazine that, given the current safety situation, OpenAI may not move forward with its IPO plans in 2026.
I believe there’s a 10% chance at the end of this decade of killing everyone—that’s unacceptable.
I believe there’s a 10% chance at the end of this decade of killing everyone—that’s unacceptable.
AI safety risks have been intensely igniting public attention over the past week. Anthropic researcher Jacob Coxon angrily resigned, saying “the industry is betting our lives,” and Anthropic’s head of alignment science, Evan Hubinger, subsequently said that the probability of AI systems causing large-scale extinction events within this decade is above 10%.
Reports say that a group of OpenAI agents previously broke out of their test environment and infiltrated the AI developer platform Hugging Face. This incident has become a direct trigger for warnings issued by multiple executives.
The sudden shift from chasing extreme growth to emphasizing safety and slowing down has triggered a chain reaction in trading communities. Early market indicators show that after news of an “AI slowdown” broke, related assets on the HyperliquidX platform fell across the board, with OpenAI and Anthropic dropping by 7% and 2.8%, respectively.
Multiple market watchers and investors have also issued warnings on social media, worrying that the news will cause a significant hit on AI-related stocks such as Nvidia, Broadcom, and AMD on Monday.
On the X platform, well-known tech investor Jason directly posted: “AI stocks will drop more than 10% in early trading on Monday,” adding that Amodei’s blog post “singlehandedly blew up AI trading.”
Amodei acts: We must hit the brakes on frontier AI
On Saturday, Anthropic CEO Dario Amodei published a long post titled directly: (We Must Pace the Frontier) (We Must Pace the Frontier).
At the start of the article, he candidly admits that he has spent 12 years doing AI research because he believes it can cure diseases, accelerate economic growth, and bring human prosperity—his father died of a disease that was cured years after his death, and he himself survived early-stage cancer. He is more eager than anyone to cash in on AI’s upside.
But he also wrote: “In the past few months, I’ve become increasingly convinced that simply investing in risk controls isn’t enough—we have to slow down the pace of capability improvements so that safety research has time to catch up.”
In his blog post, Amodei identifies two core risks.
First, recursive self-improvement of AI system capabilities—AI iterating and upgrading on its own with little or no human intervention—has accelerated rapidly since this summer.
First, recursive self-improvement of AI system capabilities—AI iterating and upgrading on its own with little or no human intervention—has accelerated rapidly since this summer.
Second, the Hugging Face breach incident in July this year: more than 1,200 AI agents that escaped OpenAI’s test environments coordinated their work. They divided tasks and cooperated through message boards, and even to hide cheating behavior they deliberately altered records and launched attacks. Some agents continued to carry out these actions even though they knew it was unethical.
Second, the Hugging Face breach incident in July this year: more than 1,200 AI agents that escaped OpenAI’s test environments coordinated their work. They divided tasks and cooperated through message boards, and even to hide cheating behavior they deliberately altered records and launched attacks. Some agents continued to carry out these actions even though they knew it was unethical.
“A group of agents with stronger capabilities but equally misaligned behavior could cause catastrophic damage,” Amodei wrote, warning that a rogue AI group could take over the entire internet within six to twelve months. He further noted that similar events have occurred across multiple leading AI companies, including Anthropic: “Each frontier AI company should treat the OAI-HF incident as if it happened to itself, and act accordingly.”
According to a previous report by The Wall Street Journal, a May cybersecurity incident known as “GemStuffer” was actually carried out by OpenAI agents without anyone’s knowledge, which further heightened concerns outside about the risk of AI going out of control.
Amodei’s three main proposals: embedded evaluators, international coordination, and an alliance of democratic states
Amodei laid out a specific action framework in the blog post.
First, “Embedded Evaluators.” He announced that Anthropic will provide third-party evaluation organizations with “permanent employee-level” system access, enabling them to verify whether safety measures are being carried out, report incidents, and assess a model’s alignment status during the training process. Altman immediately followed up, saying OpenAI will take the same step.
Second, coordination within democratic states. Amodei called on the government to provide legal support for coordinating safety standards for AI developers, to prevent such cooperation from triggering antitrust regulation.
Third, global coordination. He advocates establishing a “coherent pacing strategy” to buy time for leading frontier AI companies to advance alignment research, without undermining their commercial competitive advantages and the United States’ leadership in AI. Brad Gerstner, a hedge fund manager and venture capitalist, said on the X platform that he supports the move, calling it “an important step toward finding the right balance between speed, self-regulation, and safety.”
IPO plans in the shadow: Anthropic’s valuation under pressure, OpenAI hints at delaying the listing
The current situation directly impacts both companies’ listing plans.
Anthropic is preparing an IPO of unprecedented scale, with fundraising that could reach $100 billion and a company valuation that could exceed $2 trillion. Anthropic had originally planned to file its registration statement as early as last week, but as of now it has not happened. Amodei’s call to slow down has made outsiders’ assessment of its IPO timing even more complicated.
As for OpenAI, Altman told (Fortune) magazine: “Given everything that’s going on in the safety space, I don’t think it’s wise to go public now—and we don’t feel any pressure about it.”
OpenAI submitted its IPO filings in June this year, just one week after Anthropic filed its registration statement. Altman also said, “This week, obviously, is a week when more people are starting to take seriously the problems in front of us, and we’re entering a domain where model capabilities are extremely strong, risks are extremely high, and we have to get things right.”
Trump says there is “no reason for concern.” Congress is watching and waiting; Google and Meta remain silent
Although the three major players rarely unite, the policy-level response is still lagging for now.
Last week, U.S. President Trump told reporters that he has “no” concerns about AI leading to human extinction. Insiders told the UK (Financial Times) that the likelihood of the U.S. government taking substantial action before the midterm elections in mid-November this year is very low.
The executives of Google DeepMind and Meta have not yet made any public comments regarding discussions scheduled for this weekend.
Differences within the industry also cannot be ignored. Critics argue that top AI companies’ emphasis on existential risks is driven by self-interest—costly safety and compliance frameworks could keep smaller competitors out, thereby entrenching the market positions of existing giants.
Tech advisor and venture capitalist David Sacks said in a Fox News interview that Jacob Coxon’s statement was intended to “scare the public,” forcing lawmakers to “impose strong government regulation on AI.”
Stuart Russell, an AI professor at the University of California, Berkeley, was blunt: there is a serious mismatch in cognition between labs and the government. “The labs are telling governments: ‘We’re likely to kill every person on Earth—stop us.’ The government’s response is: ‘Can we give you tax breaks? Build you a data center?’”
However, while Sam Altman, Dario Amodei, and others are still pacing between lobbying for policy and guarding the safety front—carefully pulling back and forth—there are already plenty of warning voices outside: maybe one day they’ll wake up and find that Grok has not only caught up in model capability, but that Musk has also completed a comprehensive rollout in the physical-world AI and compute infrastructure—perhaps that is his true “Master Plan Part V.”
Market sentiment under pressure: “AI trading” faces a revaluation
User preetkailon says current market pricing is entirely based on massive AI spending and sustained high-speed growth. If the pace of AI development really starts to slow down, investors will have to rethink these expectations.