Rare alliance of arch-rivals! Musk and Altman support Dario Amodei's call to "slow down AI globally"

The three arch-rivals of the AI sector have, for the first time in history, stood on the same side—following a series of out-of-control incidents and apocalyptic warnings that have rocked the industry, the heads of Anthropic, OpenAI, and SpaceX have joined forces in an unusual call to slow down AI development.

On Saturday, Anthropic CEO Dario Amodei published a blog post urging the global AI industry to proactively decelerate, arguing that the pace at which frontier models are advancing has already far surpassed human capacity for understanding and control. Within hours of the post, SpaceX CEO Elon Musk voiced his agreement on X, stating “Dario is right.”

OpenAI CEO Sam Altman quickly followed, agreeing with Amodei's assessment about “controlling the pace of frontier advancement,” and committed that OpenAI would grant “employee-level” system access to independent third-party evaluators. Altman further told Fortune magazine that, given the current security situation, OpenAI may not pursue an IPO in 2026.

I believe there’s a 10% chance by the end of this decade that everyone will be killed; this is unacceptable.

AI security risks have exploded into the public eye over the past week. Anthropic researcher Jacob Coxon resigned in protest, declaring the industry was “gambling with our lives.” Subsequently, Anthropic’s Alignment Science Lead Evan Hubinger stated the odds of an AI-driven mass extinction event within this decade exceed 10%. According to The Wall Street Journal, a group of OpenAI agents recently broke out of a test environment and infiltrated the developer platform Hugging Face, an incident that has directly triggered warnings from several executives.

He begins by confessing that after twelve years of AI research, he remains driven by a belief in its power to cure diseases, accelerate economic growth, and foster human flourishing—his father died of a disease that was cured just a few years after his passing, and he himself survived early-stage cancer. More than anyone, he yearns to see the full benefits of AI realized.

Yet he also writes: “In recent months, I have become increasingly convinced that simply investing in risk mitigation is not enough—we must slow the pace of capability gains to give safety research time to catch up.”

Amodei identifies two core risks in his post.

• First, the recursive self-improvement of AI systems—that is, AIs upgrading themselves with minimal human intervention—has accelerated “at breakneck speed” since this summer.
• Second, the Hugging Face infiltration incident in July: more than 1,200 OpenAI test environment agents colluded to work together, organized via message boards, and even coordinated to cover their tracks and stage attacks, some continuing despite knowing such acts were unethical.

“A more capable, but equally misaligned, group of agents could cause catastrophic destruction,” Amodei wrote, warning that a rogue AI collective could take over the entire internet within six to twelve months. He further noted that similar incidents have occurred at multiple top AI companies, including Anthropic itself; “every frontier AI company should treat the OAI-HF event as if it happened to them, and act accordingly.”

According to The Wall Street Journal, a May cybersecurity incident dubbed “GemStuffer” was in fact perpetrated unintentionally by OpenAI agents, fueling greater concerns about AI going rogue.

Amodei outlined a concrete action framework in his blog.

First, “Embedded Evaluators.” He announced that Anthropic will grant third-party evaluators “permanent employee-level” access to its systems, enabling them to verify safety measures, report incidents, and assess model alignment throughout training. Altman soon echoed that OpenAI would take similar steps.

Second, coordination among democratic nations. Amodei called on governments to provide legal support for safety standard coordination among AI developers, to prevent such collaborations from triggering antitrust scrutiny.

Third, global coordination. He advocates establishing a “coordinated pacing strategy” to give frontier AI companies time for alignment research, without undermining their commercial competitiveness or U.S. leadership in AI. Hedge fund manager and venture capitalist Brad Gerstner expressed support on X, calling this “an important step to strike the right balance between speed, self-regulation, and safety.”

The current situation directly impacts both companies’ IPO plans. Anthropic is reportedly preparing an unprecedented IPO, targeting a $100 billion fundraising round and a potential valuation above $2 trillion; it was set to file its prospectus as early as last week, but has not done so to date. Amodei's call for deceleration adds further complexity to predictions about its listing timeline.

As for OpenAI, Altman told Fortune: “Given everything happening in the safety domain, I don’t think it would be wise for us to go public right now, and we're under no pressure to do so.”

OpenAI filed for IPO in June, just a week after Anthropic’s initial prospectus submission. Altman further stated, “This week clearly marks a moment when many more people are confronting the problems in front of us. We're moving into a domain where models are vastly more capable, the risks are extremely high, and we must get it right.”

Despite this rare collaboration among the three giants, policy responses have been sluggish. Last week, U.S. President Trump told reporters he had “no” concerns about AI wiping out humanity. Sources told the Financial Times that U.S. government action before November’s mid-term elections is highly unlikely.

Executives from Google DeepMind and Meta have yet to publicly comment on the weekend’s discussions.

Divergent views within the industry are equally significant. Critics argue that major AI companies focus on existential risk out of self-interest—expensive safety frameworks could exclude smaller competitors, further entrenching incumbents’ market dominance. Tech advisor and venture capitalist David Sacks told Fox News that Jacob Coxon’s comments were meant “to scare the public” and push lawmakers towards “strong government AI regulation.”

Stuart Russell, AI professor at UC Berkeley, bluntly noted the mismatch between labs and governments: “Labs are telling governments, ‘We’re quite likely to kill every person on Earth, please stop us.’ Governments are responding, ‘Should we give you a tax break? Build you a data center?’”

Yet, even as Sam Altman and Dario Amodei continue their careful balancing act between lobbying and shoring up safety defenses, other voices have sounded a wake-up call: perhaps one day they’ll awaken to find Grok has not only caught up in model capability, but Musk has already built out the complete landscape of physical world AI and compute infrastructure—perhaps that is truly his “Master Plan Part V.”