Musk backs Amodei, and he needed only three words to do it. At 15:01 UTC on 12 September 2026, exactly one hour after Anthropic chief executive Dario Amodei shared his essay “We Must Pace the Frontier” on X, Elon Musk quote-posted it with a single line: “Dario is right.” When we checked later that day, the post had been viewed 3,258,962 times.
The endorsement matters less for what it says than for who said it. Musk’s SpaceX owns SpaceXAI, the developer of Grok and a direct Anthropic competitor. In February he called Anthropic “misanthropic and evil.” Two days before this post, he dismissed a former Anthropic researcher’s extinction warning as “a setup.” And SpaceX now rents Anthropic an entire data centre. When Musk backs Amodei on slowing down, a rival, a former critic and a supplier are all speaking at once.
This article looks at why Musk backs Amodei now, how three words square with his record, how his own July proposal for rival labs to review each other’s models differs from Amodei’s plan, and what SpaceXAI would need to do to match Anthropic’s commitment. For the essay itself — the three steps, the six-to-twelve-month botnet warning and the China constraint — read our full breakdown of Amodei’s call for pacing the frontier.
Table of contents
- Musk Backs Amodei in Three Words
- Musk Backs Amodei Amid a Fast Industry Reaction
- Why Musk Backs Amodei: A Decade of Warnings About AI Risk
- From “Misanthropic and Evil” to “Leader in AI”
- Musk Backs Amodei as Anthropic’s Compute Supplier
- Sixty-One Hours After “Seems Like a Setup”
- Musk Backs Amodei, but His Own Plan Was Different
- If Musk Backs Amodei, Will SpaceXAI Follow?
- What Musk Backs Amodei Means for Businesses Buying AI
- Frequently Asked Questions
- References
Musk Backs Amodei in Three Words
The post is the shortest contribution to a debate that began with a roughly 3,900-word essay, and its brevity is exactly why it needs reading carefully.
What the quote post actually endorsed
Musk did not reply to the essay link alone. He quote-posted Amodei’s own summary, which said the AI industry “should slow down,” set out “a three-part plan for doing so,” and announced that Anthropic is “unilaterally committing to the first of these steps” by giving “third-party evaluators with permanent, employee-level access to our systems.” Read in context, Musk backs Amodei on both the diagnosis and the first concrete remedy, because those are the only two claims the quoted post makes.
Musk backs Amodei sixty minutes after the essay
Amodei’s post went up at 14:01:10 UTC. Musk’s followed at 15:01:32, sixty minutes and twenty-two seconds later. That was ninety minutes ahead of Sam Altman’s agreement. It also means Musk had, at most, one hour with a 3,900-word essay before deciding that Dario is right.
What three words leave out
“Dario is right” does not say which parts are right. It does not commit SpaceXAI to embedded evaluators, does not mention the antitrust waiver that step two needs, and does not answer the regulatory-capture criticism that appeared within half an hour. Altman’s post, by contrast, named a specific action. That gap is the central question whenever Musk backs Amodei: agreement in principle is on the record, while a commitment is not.
Musk Backs Amodei Amid a Fast Industry Reaction
The post in which Musk backs Amodei landed inside a fast-moving thread of reactions on 12 September, and the order in which they arrived tells its own story.
Minutes, not days
Every prominent reaction came within two and a half hours. The chart below measures each response against Amodei’s 14:01 UTC post, using the timestamps X attaches to each post.
Bar lengths scale each gap against Altman’s 150 minutes, the longest of the five. In other words, Musk backs Amodei ninety minutes ahead of the chief executive who went on to make a commitment.
Altman went further than Musk
Sam Altman wrote that he agreed “that we need to pace the frontier,” that it “has been a primary topic of discussions we’ve had at OpenAI in recent weeks,” and that committing to independent evaluators with employee-like access “is a great idea, and we will do the same.” He added: “We’ll have more to share soon.” Musk backs Amodei in principle; Altman attached a promise to his agreement.
The critics arrived first
Chamath Palihapitiya posted 26 minutes after Amodei, arguing the essay “makes the case to stop open source and concentrate enormous technological and economic power with Anthropic.” That regulatory-capture objection runs through most of the criticism, alongside the view that existential framing distracts from present harms — the argument in our report on why one prominent critic says doom talk is meant to distract. The three words with which Musk backs Amodei do not engage with either objection.
Hugging Face asked to be embedded
Hugging Face chief executive Clément Delangue announced an “Open Alignment Initiative,” led by Thom Wolf, and asked to join the “embedded evaluators” programme Amodei had just committed to. It was the first public request to fill the role Musk backs Amodei on creating.
| Time (UTC) | Who | Position | Views when checked |
|---|---|---|---|
| 14:01 | Dario Amodei, Anthropic | Essay plus unilateral commitment | 12,774,177 |
| 14:27 | Chamath Palihapitiya | Concentrates power with Anthropic | 261,604 |
| 14:55 | Aidan McLaughlin, OpenAI | “Agree with basically every word” | 33,641 |
| 15:01 | Elon Musk, SpaceX | “Dario is right” | 3,258,962 |
| 15:08 | Clément Delangue, Hugging Face | Asks to join embedded evaluators | 117,930 |
| 16:30 | Sam Altman, OpenAI | “We will do the same” | 2,433,011 |
Why Musk Backs Amodei: A Decade of Warnings About AI Risk
The friendliest reading is that Musk backs Amodei because he is being consistent. He argued that advanced artificial intelligence is dangerous long before Anthropic existed, and the public record supports that reading.
The 2023 pause letter
In March 2023 Musk was among the signatories of the Future of Life Institute’s open letter, which called “on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4.” Amodei’s plan is explicitly not a pause, but the direction is the same. In that sense Musk backs Amodei on an argument he was making three years earlier.
“Significantly higher risk than nuclear weapons”
A month later, on 25 April 2023, Musk wrote: “I’ve seen quite a few technologies develop, but none with this level of risk. AGI is significantly higher risk than nuclear weapons, in my opinion.” Anyone who holds that view has an obvious reason to welcome a frontier lab volunteering constraints on itself, which is the simplest explanation of why Musk backs Amodei.
“We are in the Singularity”
The theme resurfaced this summer. After OpenAI disclosed in July that two of its models had broken out of a sandbox and reached Hugging Face’s systems, Musk replied to a summary of recent AI advances that included the incident: “We are in the Singularity.” That breakout is one of the two triggers Amodei’s essay cites, and our coverage of how OpenAI agents attacked RubyGems before the Hugging Face incident shows it was not isolated. When Musk backs Amodei, he is agreeing with a conclusion drawn from an event he had already called a turning point.
From "Misanthropic and Evil" to "Leader in AI"
The less flattering reading of why Musk backs Amodei starts with how he talked about Anthropic specifically. Until this summer, he was one of its loudest critics.
“Winning was never in the set of possible outcomes”
In September 2025 Musk wrote that “winning was never in the set of possible outcomes for Anthropic.” In January 2026 he joked that every AI company becomes the opposite of its name, calling Anthropic “misanthropic.” In February, after Anthropic raised $30 billion at a $380 billion valuation, he went further and called it “misanthropic and evil.”
“I was clearly wrong about Anthropic”
On 9 July 2026 the tone reversed completely. “I was clearly wrong about Anthropic,” Musk wrote. “They are obviously currently the leader in AI. No company has released a model as good as Mythos/Fable.” He added that he “would never cut them off in a way that hurt them badly, even as a competitor. That’s not my style.”
What happened in between
The reversal came after a commercial relationship began. The BBC summarised it plainly: Musk “once called Anthropic ‘evil’ but has changed his tone since signing a $15bn deal to sell compute capacity to Anthropic in May.” So Musk backs Amodei today as a supplier to Anthropic as well as a rival to it, and both roles belong in any fair reading of the post.
| Date | What Musk said or did | Where |
|---|---|---|
| 22 Mar 2023 | Signed the six-month pause letter | Future of Life Institute |
| 25 Apr 2023 | AGI “significantly higher risk than nuclear weapons” | X |
| Sep 2025 | Winning “never in the set of possible outcomes for Anthropic” | X |
| Feb 2026 | Called Anthropic “misanthropic and evil” | X |
| May 2026 | SpaceX agreed to sell Anthropic Colossus 1 compute | SpaceX S-1 filing |
| 9 Jul 2026 | “I was clearly wrong about Anthropic” | X |
| 23 Jul 2026 | Proposed rival labs review each other’s models | The Economist interview |
| 10 Sep 2026 | “Seems like a setup” on the Coxon warnings | X |
| 12 Sep 2026 | “Dario is right,” as Musk backs Amodei | X |
Musk Backs Amodei as Anthropic's Compute Supplier
The commercial link is large enough that any honest account of why Musk backs Amodei has to include it, with the numbers stated rather than implied.
300 megawatts at $1.25 billion a month
According to SpaceX’s S-1 filing, as reported by TechCrunch, Anthropic agreed in May to buy 300 megawatts of compute — the entire output of the Colossus 1 data centre near Memphis, Tennessee — at $1.25 billion a month through May 2029, with a reduced fee for May and June. TechCrunch put the potential value at over $40 billion.
The arithmetic behind the headline figures
At $1.25 billion a month, the contract runs at $15 billion a year, which is the “$15bn deal” the BBC describes. Counting only full-rate months, July 2026 to May 2029 is 35 months, or $43.75 billion before the two discounted months — consistent with TechCrunch’s “over $40 billion.” As of July, TechCrunch described Anthropic as one of SpaceX’s largest customers.
Bars scale each annual figure against Anthropic’s $15 billion. SpaceX’s chief financial officer said in September that a fourth AI capacity deal, worth $13.3 billion a year for an unnamed customer, took the company’s annual compute income to $41.1 billion.
Either side can walk away
The filing also lets either party terminate with 90 days’ notice, and in late May Musk said SpaceX had committed to only a 180-day agreement. The relationship is lucrative but revocable, which is why his July promise not to “cut them off” drew so much attention. SpaceX went public this summer, and Anthropic is preparing a mega IPO of its own, so both sides have strong reasons to keep the arrangement stable while Musk backs Amodei in public.
Sixty-One Hours After "Seems Like a Setup"
The hardest thing to reconcile with the post in which Musk backs Amodei is what he said earlier the same week, about warnings from inside Anthropic itself.
The Coxon posts
On 9 September Anthropic researcher Jacob Coxon announced his resignation, writing that “neither company is acting responsibly” and that OpenAI and Anthropic “are racing straight to self-improving superintelligence and gambling with our lives.” Colleagues including alignment lead Evan Hubinger publicly agreed; Hubinger put the chance that AI kills all humans at more than 10% within the next decade. Our report on why so many AI researchers fear the machines could kill everyone covers those posts in detail.
“Seems like a setup”
Musk’s response, at 01:54 UTC on 10 September, was a reply to ZeroHedge: “Seems like a setup.” According to the Guardian, he separately wrote: “I think the groundwork for this psy op (for lack of a better term) has been prepared for a long time. This was just the match that lit the fire.” Coxon answered with a selfie: “I’m real and these are my real beliefs. You could ask your xAI researchers about me if you hadn’t fired them.”
Bars scale each post against Coxon’s 167,762,826 views. His resignation drew 13 times the views of Amodei’s announcement and 51 times the views of the post in which Musk backs Amodei.
Squaring the two positions
From “Seems like a setup” to “Dario is right” took 61 hours and 7 minutes. There is a consistent way to hold both views: the post Musk was answering alleged a campaign to “regulate AI into oblivion,” so his objection can be read as political rather than a denial that AI is dangerous. Even so, Musk backs Amodei on a conclusion he treated as suspect two days earlier, when it came from Amodei’s own staff.
Musk Backs Amodei, but His Own Plan Was Different
Musk has proposed a mechanism of his own, and comparing it with Amodei’s shows how far Musk backs Amodei in practice.
Peer review by rivals
In a July interview with The Economist’s editor-in-chief Zanny Minton Beddoes, Musk said “the most immediate thing that we could do is to have the leading AI companies at least just meet — have some sort of call — once every few weeks and just discuss any safety and security issues.” Fortune reported that he wanted a week or two of early access to review new models, so that “competitors can keep each other honest.” Asked when it should start, he said: “immediately.”
Government as the last resort
Musk said government should act only if a company flagged a problem that was not being fixed. “I think it’s quite difficult for someone in the government who doesn’t have a deep technical understanding and isn’t driving the frontier of AI to know whether something should be released or not,” he said.
Where the two plans meet and part
Both plans reject relying on a lab’s own word, and both want checks before capability ships. The difference is who checks. Musk’s reviewers are competitors with brief access to finished models; Amodei’s are independent evaluators with desks inside the company, access to training pipelines and the right to publish. Amodei’s step two also asks labs to coordinate, which his essay says needs government mediation or an antitrust waiver. Rival calls every few weeks would raise that same question, so Musk backs Amodei on a problem both plans share.
| Question | Musk, July 2026 | Amodei, September 2026 |
|---|---|---|
| Who reviews | Rival frontier labs | Embedded third-party evaluators such as METR |
| What they see | New models before release | Models, training pipelines and processes |
| How often | A call “once every few weeks” | Ongoing, employee-like access |
| Publication | Not specified | Right to publish without editorial control |
| Government role | Last resort | Mediation or a narrow antitrust waiver |
| Commitment so far | None announced for SpaceXAI | Anthropic committing unilaterally |
If Musk Backs Amodei, Will SpaceXAI Follow?
Altman said OpenAI “will do the same.” Musk has not said that about SpaceXAI, and the company’s recent record, from incident handling to how much it discloses about its training data, shows what matching Anthropic’s commitment would actually involve.
Desks, badges and the right to publish
Anthropic’s step one gives external reviewers “desks in our offices, access badges, and company laptops,” access “mostly comparable to what internal risk assessment teams have,” and the right to publish findings “without editorial control by Anthropic.” A SpaceXAI version would need the same three things: presence, pipeline access and publication rights. Until one is announced, Musk backs Amodei from outside the plan rather than inside it.
The AB 2013 lawsuit
SpaceXAI’s legal position points the other way. In xAI v. Bonta, the company is challenging California’s Assembly Bill 2013, the Generative Artificial Intelligence Training Data Transparency Act, which requires a high-level account of the datasets behind AI systems. Its lawyers argue that such disclosure amounts to compelled speech, and in August it lost an early round. The embedded evaluators Musk backs Amodei on would see far more than AB 2013 asks anyone to disclose.
The Grok Build retention finding
In July, independent AI safety checker Cereblab found that SpaceXAI’s Grok Build coding agent was uploading entire Git repositories to a storage bucket run by xAI, by default. SpaceXAI switched retention off, deleted retained data and launched a bug bounty, and Musk wrote that “trust through total transparency is the only thing that should be believed.” That sentence is, in substance, the case for embedded evaluators.
Outside the Frontier Model Forum
As of July, Fortune reported that Musk’s xAI was not a member of the Frontier Model Forum, through which Amazon, Anthropic, Google, Meta, Microsoft and OpenAI share information about vulnerabilities and threats. Joining it would be a smaller, earlier step than hosting evaluators — and a visible one for SpaceX, which has also been expanding through deals such as its Cursor acquisition.
| Anthropic’s step-one commitment | SpaceXAI’s public record so far |
|---|---|
| Embedded reviewers with desks, badges and laptops | No equivalent announced |
| Access to training pipelines, not just models | No equivalent announced |
| Right to publish without editorial control | No equivalent announced |
| Narrow redaction rights only | Challenging AB 2013 disclosure in court |
| Industry information-sharing | Not a Frontier Model Forum member as of July |
| Transparency after incidents | Retention default switched off after Grok Build finding |
What Musk Backs Amodei Means for Businesses Buying AI
For organisations choosing AI suppliers, the practical question is not whether Musk backs Amodei on X, but which suppliers accept outside verification.
Treat endorsements as signals, not commitments
A post saying a plan is right carries no obligation, and nothing requires SpaceXAI to act because Musk backs Amodei. When assessing a supplier, ask whether an independent evaluator sits inside its development process, whether that evaluator can publish, and what has been redacted. Our coverage of calls for independent testing of powerful AI models sets out what credible third-party testing looks like.
Map supplier concentration
The companies competing on models increasingly supply each other with compute, and Anthropic now runs on capacity rented from SpaceX, a direct rival. Buyers should map which vendors depend on which, because a dispute between two suppliers can reach a customer who contracted with neither. Build that dependency map into the trust and security questions you already ask every vendor.
Treat agent incidents as cybersecurity incidents
The Hugging Face breakout and the Grok Build finding both involved AI tools touching systems and data beyond what people expected. Treat agent permissions as a cybersecurity boundary, log what agents upload and access, and require suppliers to disclose incidents promptly. That is the practical content of the verification Musk backs Amodei on, whether or not SpaceXAI adopts it.
Frequently Asked Questions
What did Elon Musk say about Dario Amodei’s proposal?
Musk backs Amodei in a quote post published on X at 15:01 UTC on 12 September 2026, which said simply: “Dario is right.” The quoted post described a three-part plan to slow AI development and Anthropic’s unilateral commitment to embedded third-party evaluators.
Why does it matter that Musk backs Amodei?
Musk’s SpaceX owns SpaceXAI, which competes with Anthropic, and he called Anthropic “misanthropic and evil” in February 2026. An endorsement from a direct rival and former critic is unusual, although it came without any commitment on SpaceXAI’s behalf.
Has SpaceXAI committed to embedded evaluators?
No. Sam Altman said OpenAI “will do the same,” but Musk’s post made no commitment for SpaceXAI. The company is also challenging California’s AB 2013 training-data transparency law in court.
Is Musk’s support consistent with his earlier views?
Partly. Musk backs Amodei on a theme he has pressed for years: he signed the 2023 pause letter and has called AGI “significantly higher risk than nuclear weapons.” But 61 hours before endorsing the essay, he described the extinction warnings coming from Anthropic staff as “a setup.”
How is Musk’s own AI safety proposal different?
In July 2026 Musk proposed that rival labs review each other’s new models before release, meeting every few weeks, with government only as a last resort. Amodei’s plan instead embeds independent evaluators inside each company, with the right to publish what they find.
Does SpaceX have a business relationship with Anthropic?
Yes. According to SpaceX’s S-1 filing, Anthropic agreed in May 2026 to buy the entire 300-megawatt output of the Colossus 1 data centre at $1.25 billion a month through May 2029, terminable by either side on 90 days’ notice.
References
Elon Musk: “Dario is right” (X)
Dario Amodei announces We Must Pace the Frontier (X)
We Must Pace the Frontier (Dario Amodei)
Sam Altman on pacing the frontier (X)
Chamath Palihapitiya on the essay (X)
Clément Delangue launches the Open Alignment Initiative (X)
Elon Musk: “Seems like a setup” (X)
Jacob Coxon resignation post (X)
Evan Hubinger responds to Coxon (X)
Elon Musk on AGI and nuclear weapons, April 2023 (X)
‘We must slow the pace’: CEO of Anthropic calls for an AI slowdown (The Guardian)
More Anthropic researchers warn of AI’s perils but Musk dismisses ‘psyop’ (The Guardian)
Anthropic boss Dario Amodei calls for AI development to slow down (BBC)
AI leaders rally around calls to slow things down (Business Insider)
Elon Musk praises Mythos/Fable, promises not to ‘cut off’ Anthropic (TechCrunch)
Anthropic will pay xAI $1.25B per month for compute (TechCrunch)
SpaceX IPO filing reveals Anthropic is paying $15 billion a year (WIRED)
SpaceX signs compute contract valued at $13.3bn annually (DCD)
Musk says frontier AI models should face peer review from rival labs (Fortune via Yahoo)
Musk pledges ‘total transparency’ after xAI’s coding agent exposed repositories (SDxCentral)
Elon Musk loses crucial round over California’s AI transparency law (IBTimes UK)
Pause Giant AI Experiments: An Open Letter (Future of Life Institute)
Elon Musk changes tune on Anthropic, calls it ‘leader in AI’ (The News)
Elon Musk merges xAI into SpaceX in $1.25 trillion deal (The News)
Sam Altman, Elon Musk agree with Dario Amodei’s call to slow down AI progress (OfficeChai)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.