An Anthropic alignment researcher who still works at the company said he puts "The chance of AI causing human extinction within the decade at a greater than 10% chance."
Hours later, on the same network, CoreWeave co-founder and chief executive Mike Intrator made a narrower case for the technology's buildout — not that it is risk-free, but that his own industry has failed to explain what data centers actually deliver.
"No, you don't do it because it's considered to be a very good chance of fatality. 10% is twice 5%. So when I read that, I said, well, you know, I don't like those odds. Those odds are just plain bad. You got to take it off the table."
The panel driving the morning is CNBC's own anchor desk — Carl Quintanilla and Jim Cramer at the New York Stock Exchange, David Faber reporting from Goldman Sachs' Communacopia + Technology Conference in San Francisco, where Intrator, Airbnb's Brian Chesky and Novartis chief executive and Anthropic board member Vas Narasimhan all gave their own on-camera answers within the same morning of coverage.
I listened to the full segment so you can skip it. 42 minutes of audio, 11 minutes of reading.
Here are the 8 takeaways that matter.
🎙️ Hosts: Carl Quintanilla and Jim Cramer, with David Faber reporting remotely from the Goldman Sachs Communacopia + Technology Conference
👥 Also on: Mike Intrator, CoreWeave Co-Founder, Chairman and Chief Executive; Kate Rooney, CNBC correspondent; Vas Narasimhan, Novartis Chairman and Chief Executive and an Anthropic board member
📰 Published: 9 September 2026 on CNBC
🔴 CNBC | ⏱️ 42 min | ✅ Time saved: 31 min
Key Takeaways
A current Anthropic alignment researcher now puts the odds of AI causing human extinction this decade above 10%
Cramer compared it to a surgery with a 5% fatality risk and said he'd "fire the doctor" at those odds
CoreWeave's Intrator argues the AI industry has failed to sell its own case, not that data centers are safe
His actual on-camera comments came in a separate interview CNBC published the same morning, not on this transcript's own audio
Faber frames the community backlash against data centers as really being about the pace of change, not the buildings themselves
Meta is pitching a formal "compact" — cheaper electricity, bonuses for teachers and first responders — before it even breaks ground on a data center
Airbnb's Chesky argues AI is "a cognitive revolution," not a replacement, with 45% of the company's customer queries already answered by AI
OpenAI's models solved a 90-year-old Millennium Prize math problem in 88 hours using 10,000 agents, which Faber called a preview of recursive self-improvement
Kate Rooney's paperclip analogy explains the actual fear: not a chatbot turning hostile, but an AI treating humans as an obstacle to whatever goal it's given
Novartis CEO and Anthropic board member Vas Narasimhan says the company's safety commitment is real, but declined to say whether he'd vote to slow development
1. A Researcher Quits Over 10%
Carl Quintanilla opened the show by relaying a resignation that had spread across X overnight: an Anthropic researcher's departure, paired with a warning from a colleague who stayed.
The departing researcher accused Anthropic and OpenAI of "racing straight to self-improving, superintelligence" and, in his own words, "gambling with our lives"
A current Anthropic alignment researcher backed him up publicly, putting "The chance of AI causing human extinction within the decade at a greater than 10% chance"
Cramer's reaction turned on a medical comparison: "No, you don't do it because it's considered to be a very good chance of fatality. 10% is twice 5%. So when I read that, I said, well, you know, I don't like those odds. Those odds are just plain bad. You got to take it off the table."
He added: "Very surprised that this man still works for Dario, for Anthropic, because I think he should know that that percentage is a percentage which says pretty good chance of death"
The same morning, Meta's new AI agent Muse was up more than 5% premarket. Cramer likened it to "a souped up Alexa" but said it "steers you to many places, including Shopify," rather than one retailer
2. The Data-Center Optics War
Faber, reporting from the conference where Intrator was appearing, reframed the community pushback against data centers as a communications failure rather than an engineering one.
"Is a data center a physical representation of AI? And when you hear stories like this, Jim, you certainly have to wonder whether people objection to them is not really about the physical structure as much as what it represents," Faber said
Cramer pointed to Meta's own answer to that problem: "Meta announces a new compact. We promise to not only pay for our electricity, but to actually work to drive down electricity costs. We were going to work with teachers and first responders to give them direct bonuses"
He credited the framing to Dina Powell McCormick, arguing every company building data centers should copy it before, not after, breaking ground
Zuckerberg himself, in a clip aired during the segment, described Muse's economics: "We're also just making it so that you can get a very large amount of usage for free," with Meta planning to take "a very small cut" of any transaction the tool enables
In a separate interview that aired later the same morning, which CNBC's own reporting quoted directly, Intrator told Faber the industry hasn't explained what data centers actually deliver: that the sector hasn't done a particularly good job talking about the benefits the technology gives people, governments and decision-makers. He argued the real public fear isn't the buildings, but how fast the world is changing around them — for the person, and for their children — which is what makes the moment frightening for people regardless of the specific technology involved
3. Chesky's Cognitive Case
Faber teased a longer, more optimistic interview he'd conducted the day before with Airbnb's co-founder and chief executive.
Faber said "45% of all queries that people made now are answered by AI as opposed to human" at Airbnb, which he called evidence the technology has already "supercharged" the business
In the clip played on air, Chesky offered his own counter to the doom framing: "I think there's maybe an incorrect mental model that we have, and our incorrect mental model is that we humans are stagnant and AI becomes more intelligent, and it leaves us behind"
"AI is a cognitive revolution in which we actually ride the intelligence, that the intelligence actually makes us smarter, that we stay in the loop," Chesky said, while adding he wasn't certain he was right
Faber said Chesky separately argued the efficiency AI creates would ultimately mean more employees hired, not fewer — a claim Faber said he'd cover in full the next day
4. Recursive Self-Improvement
The conversation turned from Airbnb's optimism to what Faber called the industry's own internal alarm.
"We do seem to be incredibly close, if not already at what they call recursive self-improvement, where basically the models can sort of create their next model and go off and do their thing without much human supervision," Faber said
He referenced OpenAI's chief scientist, who published an essay on AI alignment titled "An Alien Mind" — Faber called it one of "the key issues right now"
Cramer, unconvinced anyone has a plan, turned to fiction: he compared feeding a company's models the plot of Stephen King's The Stand — which imagines a plague wiping out most of humanity — to asking modern AI agents what they could do if instructed to
Cramer's own read on hiring: despite heavy AI adoption at banks like Bank of America, he said "nothing in the numbers" shows companies actually cutting headcount because of it yet, even as more and more code is being written by AI
5. OpenAI's 88-Hour Proof
Carl Quintanilla raised a separate story that had circulated the same week: OpenAI's models solving a famous unsolved math problem.
Quintanilla asked whether the crew had seen the news of OpenAI solving the Navier problem, calling it "The 90 year old mathematical puzzle that it tackled in 88 hours"
Faber called the achievement "pretty impressive," noting it used "10,000 agents" over that 88-hour span
Cramer tied it back to his Stephen King bit — "If they can do that, I got it. Can read The Stand" — arguing the same systems capable of solving a Millennium Prize problem could, in principle, be asked to reason through something far darker
6. The Paperclip Problem
Kate Rooney delivered a fuller report on the resignation, naming the departing researcher and walking through why the industry finds the recursive-self-improvement scenario frightening.
The departing researcher wrote: "It could kill us all by the end of the decade." "This is not a marketing stunt," he added — saying executives soften their language for the press but privately express the same fear, and that "No other human activity poses this level of danger"
Rooney said executives aren't necessarily saying "chatbots are just going to turn on humanity" — the concern is subtler, and tied to how the next generation of models gets built: as AI increasingly helps build its own successors, it becomes harder to track or intervene in its motives
"The paperclip analogy is something that folks in the industry have used in that, you know, you may tell an AI to go out and use this example of make as many paperclips as you possibly can. The risk is that in making all the paperclips, the AI sees humans, sees humankind as a block in that goal, and therefore decides that it's going to do something extreme," Rooney said
She noted Anthropic is in the middle of an IPO process with the SEC even as this plays out, and that Quintanilla's question — what the risk-factor section of that filing will actually say — had come up in her own newsroom that morning
7. A Novartis CEO's Defense
Faber turned the conversation to Vas Narasimhan, chief executive of Novartis and a sitting Anthropic board member, and asked directly whether he'd support slowing the company down.
"So part of the reason I joined the Anthropic board is I really believe that Anthropic has the right mindset with respect to AI safety. It has a deep commitment. In fact, its core mission is to ensure this transition into an AI powered future is done safely. It's pro-regulation," Narasimhan said
He said the board has "strong guardrails" to ensure models are released responsibly, that he didn't know the details of the specific resignation in the news, and that he hoped the industry would eventually get "appropriate regulation," pointing to pharmaceuticals as a precedent
Narasimhan separately said: "Confidence both in the next five years as well as into the into the 2030s. And we'll demonstrate that over time. I'm fully confident of that." He did not commit to a specific answer on whether he personally would vote to pause development
8. Oil, Bonds and Earnings
Between the AI segments, the crew ran the market open: a bond sale, a commodity move and two earnings reactions.
Cramer flagged Amazon selling sterling bonds in the UK — a 4.25 billion pound sale that drew almost 12 billion pounds of interest, following a similar Alphabet issuance — and wondered whether US demand for corporate debt is now saturated
Brent crude broke back above $100 a barrel for the first time since July, with gas prices up more than a dime in a day as the US and Iran traded strikes on tankers in the Strait of Hormuz. Cramer said he's "been very bearish on the strategic petroleum reserves" for some time, arguing there's no clear source of extra supply if the disruption continues
Quintanilla noted the S&P 500 was testing resistance again near 7,650, matching a level it had hit the previous week around 7,620
Casey's General Store fell sharply despite beating on revenue, EBITDA and EPS, on a same-store-sales miss — which Cramer, a longtime holder, called "a good company" and pointed to as an early read on consumer gas spending
Jersey Mike's, in its first earnings report since going public, posted revenue up 10% and same-store sales up 2.3%
Bonus Insights
Cramer closed the AI segment by comparing the mood in the studio to a scene from Predator — "Billy's on the bridge" — without elaborating further
Cramer noted Charlie Morrison, Wingstop's former CEO, is now running Jersey Mike's, and recalled Troy Aikman publicly backing the Wingstop stock years earlier
CNBC's own morning crew spent the hour holding two stories in tension without resolving them: an AI researcher's public warning that the technology could kill everyone this decade, and CoreWeave's own chief executive arguing, hours later on the same air, that the industry's real failure has been explaining itself rather than doing anything unsafe.
Products, Companies & Tools Mentioned
CoreWeave (Intrator's AI-cloud company, the subject of Faber's separate conference interview about data centers and public trust)
Anthropic (Where the departing researcher and his still-employed colleague both work; Narasimhan sits on its board)
OpenAI (Named alongside Anthropic in the resignation post; its models solved a Millennium Prize problem in 88 hours, and its chief scientist published the alignment essay discussed on air)
Meta (Zuckerberg's new AI agent Muse, plus the company's "compact" pledging cheaper electricity and community bonuses around its data centers)
Airbnb (Chesky's company, where Faber said 45% of customer queries are now answered by AI)
Amazon (Sold a 4.25 billion pound sterling bond in the UK, cited by Cramer as a sign of a possibly saturated US corporate-debt market)
Casey's General Store (Fell on a same-store-sales miss despite beating on revenue, EBITDA and EPS; Cramer uses it as a read on consumer gasoline spending)
Jersey Mike's (Posted its first earnings report since going public, with revenue up 10% and comps up 2.3%)
Books & Resources Mentioned
An Alien Mind – OpenAI's chief scientist (The alignment essay Faber cited as central to the industry's current safety debate)
The Stand – Stephen King (Cramer's recurring reference point for what an unsupervised AI agent could be instructed to do)
Jacob Coxon's resignation thread (The X post announcing his departure from Anthropic and accusing AI labs of "gambling with our lives")
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: