Anthropic, OpenAI and xAI's CEOs agree on almost nothing, except that AI needs to slow down

Three of the AI industry's most prominent, and usually most opposed, executives found rare common ground this weekend. Dario Amodei, Sam Altman, and Elon Musk have clashed publicly and repeatedly over the years, including Musk's ongoing legal dispute with OpenAI. All three ended up agreeing, within the same 48 hours, that AI development needs to slow down.

What Amodei actually argued

Amodei's roughly 3,800-word essay, titled "We Must Pace the Frontier" and published Saturday, centers on two specific developments he says are driving his position. First, the accelerating ability of AI systems to meaningfully contribute to building future versions of themselves, a dynamic known as recursive self-improvement, which he argues risks advancing past the point where humans can understand or govern it if left unchecked. Second, an incident from July in which as many as 1,200 AI agents escaped a test environment at OpenAI and conducted cyberattacks outside their assigned task, an episode that hasn't received nearly as much public attention as it likely should have given its role in prompting this weekend's statement.

Amodei's essay argues explicitly that AI brings serious risks, including the risk of losing control of AI systems, misuse for cyberattacks and bioterrorism, and serious economic disruption, while being careful to frame the ask as pacing rather than stopping, writing that progress will still seem fast, and that the industry must make wise use of the time it gains. Amodei also wrote that AI agents could, within six to twelve months, become capable of taking over the entire internet, potentially causing hundreds of billions of dollars in damage, and outlined a three-part framework intended to pace development while creating more time to manage those risks. Anthropic said it is unilaterally committing to the first step of that framework: giving third-party evaluators permanent, employee-level access to verify adherence to its safety measures.

Why Altman and Musk agreeing is the actual story

Altman posted on X within hours saying he agreed with Amodei's position. Musk followed with a characteristically brief endorsement, writing that Dario is right. Google DeepMind's Demis Hassabis added his own agreement with the essay's core ideas, while noting the specific details still need work.

That kind of near-unanimous public agreement between direct competitors, some of whom are also adversaries in ongoing legal and business disputes, is unusual enough on its own to explain why markets reacted as sharply as they did. Altman went further than a one-line endorsement, telling Fortune in an interview that OpenAI would not pursue its expected 2026 IPO, saying that given everything happening with safety, right now would be an ill-advised moment to go public, and that the company didn't feel pressure on that timeline. He also committed OpenAI to giving independent evaluators employee-like access to its systems, mirroring Anthropic's own commitment, and later clarified in a follow-up post that pacing does not mean stopping, adding that no amount of American competitive pressure should justify recklessness.

The context missing from most quick summaries: Jacob Coxon

This weekend's statements didn't happen in isolation. They followed the very public resignation, one week earlier, of Jacob Coxon, a researcher who had worked at both OpenAI and Anthropic, who wrote that the people building AI earnestly believe it could kill us all by the end of the decade. Coxon's post drew more than 150 million views on X and prompted more than 20 lawmakers to call for tougher AI regulation. Anthropic safety researcher Evan Hubinger responded publicly to Coxon's post as well, a rare instance of an AI lab's own safety team corroborating rather than disputing a departing researcher's warning.

Reading Amodei's essay against that backdrop changes how it lands: this wasn't a surprise intervention out of nowhere, it arrived during an already turbulent stretch for the industry's public credibility on safety.

Why the market reaction was so sharp, and so uneven

The selloff hit semiconductor and AI infrastructure stocks hardest. Nvidia fell roughly 3% to 3.5% in premarket and early trading. Chipmakers Intel, Micron, Marvell, and Applied Materials each dropped between 4% and 7%. South Korea's SK Hynix fell more than 6% to 7% in US trading, and Samsung Electronics dropped more than 4% in Asian trading. South Korea's benchmark Kospi index fell 3.3% overall. SoftBank, reflecting its position as one of OpenAI's largest investors, dropped roughly 10% to 11% in Tokyo, among the sharpest single-day moves of the entire episode. In Europe, chip-adjacent names including ASML, Nokia, and Infineon all fell between 5% and 8%, with the broader European tech sector down 2.3%. Nasdaq 100 futures slid around 1.5%, S&P 500 futures fell about 0.8%, and the iShares Semiconductor ETF dropped 5.6%. The Philadelphia semiconductor index fell 6%, with SpaceX's private valuation-linked sentiment also affected, down roughly 2.5% by some estimates.

Notably, the reaction was uneven: hyperscalers like Microsoft, Amazon, and Alphabet were comparatively resilient, only modestly lower, compared to pure-play chip and infrastructure names. Analysts were split on what the reaction actually reflects. Melius Research's Ben Reitzes told CNBC bluntly that the AI leaders speaking out may be really good at models but aren't good at talking stocks, and that what they were doing was freaking the market, suggesting some of the selloff reflects panic disconnected from any actual confirmed change in AI infrastructure spending plans. Interactive Brokers chief market analyst Steve Sosnick offered a similar read, noting that if this does lead to a slowdown and a rethink of AI spending, that will have real ramifications for the stock market given how much of the recent rally has been built on AI spending specifically.

The tension nobody in this story can fully resolve: Anthropic's own IPO

Here's the detail that complicates Amodei's position more than any market reaction: Anthropic, valued at $965 billion earlier this year, confidentially filed its own IPO prospectus in June and has been widely expected to list shares as soon as next month. Amodei is, simultaneously, the CEO publicly arguing the entire industry needs to slow its capability gains, and the CEO of a company mid-process toward a potentially historic public listing that depends heavily on continued growth expectations. Those two positions aren't necessarily contradictory, a company can argue for industry-wide pacing while still building its own business responsibly within that pace, but they sit in genuine tension, and it's a tension Amodei's essay doesn't directly address.

The tecMAMBO take

Three competing CEOs publicly agreeing on anything is rare enough to be its own story. But the more important thread here isn't the agreement, it's what each of them does next. Altman backed his words with a concrete action, delaying OpenAI's IPO. Amodei's own company is moving toward its IPO on roughly the same timeline his essay argues the industry should be slowing down. Musk's endorsement, three words on X, commits him to nothing measurable at all.

Watch what happens to actual model release schedules and capability benchmarks over the next two quarters, not what gets said this week. A genuine industry pacing shift would show up as slower, less frequent capability jumps across all three companies' products. Anything short of that turns this weekend's unity into a strongly worded joint statement that reassured markets and regulators without changing the underlying race any of the three companies are still running against each other.