Rob Leathern's read on a weekend of AI-extinction warnings is that the danger is many small incidents rather than one runaway system.
The industry's own leaders spent the weekend saying the technology might kill everyone. Leathern, who ran product teams at two of the companies now arguing about it, thinks the safety concerns are genuine and the timing is commercial: the same statements let a set of firms that have been outspending each other stop doing so.
"I think two things can be true."
Leathern was a product executive at Google and at Meta before co-founding InfoHawk, so the platform-safety trade-offs being debated in public are ones he has had to make internally.
The full segment is covered here so you can skip it.
Here are the 4 arguments that matter.
👤 Guest: Rob Leathern, co-founder and CEO of InfoHawk, previously a Product Executive at Google and at Meta
🎙️ Hosts: Carl Quintanilla and Sara Eisen, anchors of CNBC's Squawk on the Street
👥 Also on: Rapidan Energy's Scott Modell, Melius Research's Ben Reitzes and JPMorgan Asset Management's Jack Caffrey, in separate segments of the same hour
📰 Published: 14 September 2026 on the Squawk on the Street podcast feed (CNBC)
🟣 Apple Podcasts | 🔗 Segment page | ⏱️ length not available
Key Takeaways
He rates a runaway AI as less likely than a long run of isolated incidents
The safety and security concerns themselves he calls real
The slowdown statements are commercially useful to every firm that signed them
They have been locked in an escalating race to ship features
The gap to Chinese open-source models is what gives US labs room to cut spending
Anthropic's offer of outside access is the part he reads as a genuine cost
1. Hyperbolic, but Real
The desk brought Leathern on straight off a fresh Truth Social post from the president, which said the growing calls for an AI slowdown exist only because the United States is leading, and told the industry not to kill the golden goose. The first question put it to him that he finds the kill-us-all claim overstated.
"I think it is a bit hyperbolic."
He does not dismiss the underlying risk: "I think the safety and security concerns are definitely real. Many of them are real."
The distinction he draws is about shape rather than severity: "I think runaway AI is certainly, though less likely to occur than a lot of isolated incidents that we'll probably have to deal with over time."
"That said, you know, I think two things can be true."
His two-part conclusion is that the concerns and the positioning coexist: "We can also have those real security and safety concerns while also recognizing that some real politic going on here."
2. Why a Slowdown Helps
A host said the desk had been waiting to hear whether Google would respond directly to the president's post, and asked what Leathern expects it to say if it says anything.
He would not guess at Google's answer, and moved instead to who benefits from the slowdown language.
"I think that some of these calls for a slowdown definitely help all of these companies because they've really been locked in a battle and escalating battle with each other to ship more features."
"The teams working on these products are working very hard." In some cases, he said, "Maybe they're trying to go through the safety checks more quickly than they would otherwise."
He credits the industry on the comparison to earlier technology, saying the labs are "doing a lot more on the safety side than perhaps we have seen in previous technology rollouts" — while adding that there is now a recognition inside them that they need to do more.
3. The Chinese Gap
Asked how the United States can pace itself if China does not, Leathern gave the commercial answer twice over.
"They don't want to be stuck trying to outspend each other necessarily."
The lead over Chinese open-source models is what makes restraint affordable: while the US labs still have a presumed gap over those models, he said, they have "room to try to rein in spending" — with the hedge "perhaps" attached.
The commitments are not unconditional. He pointed to "some asterisks in these statements saying that if situations change, you know, we may have to go back to what we were doing before."
4. Anthropic's Real Give
The part of the weekend he singled out as substantive was Anthropic's, and his test for it is cost.
His rule for reading a disarmament pledge is whether it hurts: when a firm unilaterally disarms, he said, you want to be sure there is "a real cost to you" attached.
Anthropic's version clears that bar because it lets outsiders in — "having outside folks come into the company and have access to all of the models and tools that the internal teams do" is disruptive, and he said it "is a real give by the company."
Bonus Insights
The show framed his position for the audience before he said a word of it, and the framing was commercial rather than technical: the desk said its next guest believes the push to slow down AI progress is more about getting AI spending under control, after AI capital expenditure among the hyperscalers soared this year. That is the show's characterization, and Leathern neither repeated it nor disputed it.
The president's post, as the desk read it out, made the same commercial argument from the other direction: the calls for a slowdown are happening because the United States is ahead, so slowing down would kill the golden goose.
Leathern's bottom line is that the extinction framing and the spending question are the same conversation: the risks he expects are numerous and small rather than singular and final, and a coordinated slowdown is worth more to the labs as a way out of a capital-expenditure race than as a safety measure — with Anthropic's offer of outside access the one commitment that costs the company something.
Products, Companies & Tools Mentioned
InfoHawk (The company Leathern co-founded and runs, which builds AI systems to detect online deception — the reason the desk brings him on for platform-safety questions)
Anthropic (The statement he singles out as a real concession, because outside evaluators would get the access internal teams have)
Google and Meta (His two former employers; the desk was still waiting on a direct response from Google to the president's post)
Truth Social (Where the president's post about not killing the golden goose landed moments before the segment)
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: