Alex Kantrowitz said the probabilities of human extinction being circulated by current and former AI-lab employees are invented figures presented as arithmetic.
The debate has moved from research papers to politics in a week, with Paul Tudor Jones writing an op-ed on self-improving AI and Jensen Huang and Brad Gerstner publicly attacking an Anthropic researcher's comments. Kantrowitz's position is that there is real risk and that almost none of the numbers attached to it mean anything.
"Like, they're advancing these probabilities like it's rooted in science and it's rooted in math. It's not, it's numbers that are made up and just speculation based off of what's going to happen."
Kantrowitz founded Big Technology, writes the newsletter of the same name, and is a CNBC contributor who covers the large platform companies.
I listened to the full segment so you can skip it.
Here are the 4 arguments that matter.
👤 Guest: Alex Kantrowitz, founder of Big Technology and a CNBC contributor
🎙️ Host: Scott Wapner, a co-anchor of CNBC's Closing Bell
📰 Published: 10 September 2026 on CNBC's Closing Bell
🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 5 min
Key Takeaways
The 10% extinction probabilities are asserted, not calculated, on Kantrowitz's reading
His counterweight: today's systems can barely order a pizza
The story has left the industry and reached Washington and state governors, which he says makes pauses and kill switches a live legislative prospect
The timing is the worst possible for the industry, with OpenAI and Anthropic listings he says may be priced near $2 trillion
He accepts the risk and rejects the extrapolation: recursive self-improvement has no evidence behind it yet
1. Made-Up Probabilities
Scott Wapner opened by asking whether the conversation has drifted from science fact into science fiction, after CNBC's Kate Rooney reported Jensen Huang putting a much longer timeline on the technology than the industry assumed a couple of years ago.
Kantrowitz's answer was yes, and his objection is to the form of the claim rather than the subject. "Like, they're advancing these probabilities like it's rooted in science and it's rooted in math. It's not, it's numbers that are made up and just speculation based off of what's going to happen."
His reality check on current capability was deliberately mundane. "And remember, today's technology can barely order you a pizza."
He asked for proportion rather than silence. In his words, the industry needs to hold its horses on what has spun out of the debate, which is a widespread reading that AI is likely to harm or wipe out humans
2. The Story Left Containment
The reason he thinks this matters commercially is where the argument has travelled.
It is no longer an industry conversation. "And so certainly this story has sort of left containment, and it's now spread into places like Washington and governors' mansions across the United States"
That makes previously theoretical policy real. He said it could be a turning point at which pauses or kill switches are debated far more seriously than they have been so far
Wapner pointed out that some of the alarms are coming from inside the industry, at a moment when the sector is already being criticized for public distrust, and that other insiders are trying to talk the fear down
3. A Perfect Storm For AI
Kantrowitz put the political risk against the money already committed.
His summary of the position was a compounding one. "Well, on the risk side, you're looking at a perfect storm here, right?" AI was already unpopular coming into the week: people dislike data centers and fear job losses. The possibility that AI kills them or their children was not part of the discussion and now is
The consequence is that fears about government intervention stopping the build-out are now concrete rather than abstract
Against that sits the capital. He named Nvidia's market capitalization and the imminent OpenAI and Anthropic public listings, which he said look as though they might be priced at $2 trillion. "For all this to work in many ways, a lot of this needs to go perfect."
His forecast for the industry's response was unambiguous. "The industry is going to push back on this extremely hard." He read Huang's remarks and Brad Gerstner amplifying them on X — where Gerstner called an Anthropic researcher's comment outlandish, "deeply untrue" and, in his paraphrase of Huang, "wrong, arrogant" — as the start of that
4. Risk Without Extrapolation
Wapner put the strongest version of the case to him: Paul Tudor Jones is not a hysterical person, and Anthropic itself had published something alarming hours earlier.
The host read from Anthropic's own post, which said the company had published "we published our most detailed threat intelligence report to date. It covers how people tried to misuse Claude for cyberattacks, influence operations, surveillance, biology, and building weapons, and how we found and stopped them. We disrupted every operation in the report."
Kantrowitz conceded the risk without qualification. "And I'm not saying there's no risk, there's definitely risk." He named the Hugging Face incident of the past month as genuinely concerning, and the coordination of many models at once — his example was the OpenAI effort that solved a Millennium Problem — as a real capability
His request of the industry is to separate the observation from the projection. The Hugging Face incident happened; that does not, on his account, mean AI is more likely to wipe out humanity
The specific claim he says has no support is self-improvement. "recursive self-improvement hasn't proven that it's going to be the case. AI improving other AIs, we don't have any evidence that that is going to come."
Kantrowitz's bottom line is that the risk debate has become a market risk, because the policy response will be shaped by numbers that were never calculated.
Bonus Insights
The segment was set up by CNBC's Kate Rooney, reporting that Jensen Huang had put a much longer timeline on the technology than the industry would have expected a couple of years ago, which she read as him saying there is time to work the problem out
Kantrowitz's sequencing of the backlash is worth keeping. People objected to data centers first and to job losses second; the possibility of physical harm is a third objection that did not exist in the public debate until this week, and it arrives on top of the other two
Wapner noted that Brad Gerstner had been in the room for Huang's remarks and posted about them with a photograph, which is how the pushback reached a wider audience
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: