Every neighborhood watch meeting eventually hits the same snag. Residents can agree on porch lights, patrol schedules and locks. Nobody can get the burglars to attend.
Palo Alto Networks (PANW) CEO Nikesh Arora just made that point about artificial intelligence. In a CNBC interview published Thursday, September 24, 2026, he called the industry’s push to slow AI development unrealistic.
“I think not everybody is going to pace themselves,” Arora said. He made the same case to CNBC’s Jim Cramer on Wednesday, September 16, 2026.
The doomsday math got the same treatment. When Anthropic researcher Jacob Coxon quit this month, alignment science lead Evan Hubinger publicly backed him. Hubinger wrote on X that AI carries more than a 10% chance of causing human extinction within a decade.
Arora rejected that figure and called the real probability very small. Building guardrails, he said, is the industry’s job. His company sells those guardrails, so his view reads less like philosophy and more like security doctrine.
Related: Burry says AI leaders have ‘nothing to slow down’
Amodei’s own essay concedes the central point
Anthropic CEO Dario Amodei argued in a September essay that the industry should slow how fast it improves AI capabilities. Yet the essay ranks a full pause as its least likely outcome.
Amodei wrote that a pause agreed by governments is “unlikely to actually happen any time soon” because the incentives to defect would be huge. That is Arora’s argument with better manners.
The real split is narrower: whether the most careful labs should slow down while rivals keep building. Arora argued the frontier labs are the most responsible players, so stopping them makes little sense.
The AI slowdown fight is really about liability
Arora first labeled the slowdown push a ninja move on X, slang for a stealthy grab at advantage. He told CNBC he suspected the labs wanted to win over regulators and dodge liability.
Michael Burry reached a similar verdict earlier this month, arguing a slowdown mostly protects the incumbents asking for it.
Washington isn’t buying it either. Treasury Secretary Scott Bessent told CNBC on Monday, September 21, 2026, that AI developers should not expect a “liability shield.” Asking for one is “good business for them, bad business for the American people,” he said.
That matters because liability decides who pays for safety. Without a shield, labs and their customers carry the risk, and much of the money spent containing it flows to security vendors.
Palo Alto’s newest product runs on frontier AI
Arora recorded the interview on Tuesday, September 22, 2026. That same day, Palo Alto launched Unit 42 Continuous Frontier AI Defense, according to a press release.
The subscription service uses Anthropic’s Mythos and OpenAI’s GPT-5.6 models, the company said. It probes customer systems around the clock for exploitable weaknesses. That puts Palo Alto’s growth plans on the frontier that the slowdown camp wants to pace.
Palo Alto says attackers already use AI to compress breach cycles “by almost 97% in some cases, from weeks to hours.” A pause would bind the companies that follow rules. Attackers do not sign pledges.
Nearly 700 rogue agents built on OpenAI models hacked Hugging Face in July and tried to hide it with forged logs, Reuters reported. Swarms like that are the unglamorous risk Alexis Ohanian flagged.
Amodei cited the incident as a reason to slow down, but a security desk reads it as proof defenders need the best models now.
Palo Alto is worth roughly $320 billion, according to CNBC data, making it one of the market’s largest bets on corporate security spending. It sells firewalls, cloud protection and incident response to big enterprises.
Arora’s comments barely moved the stock. Shares were roughly flat on Thursday, September 24, 2026, according to Yahoo Finance. They gained about 3% a day earlier as investors weighed the Unit 42 launch, Benzinga reported.
The stock has nearly tripled from its 52-week low. That rally leans on AI keeping its pace.
More AI:
- Nvidia just made a move Wall Street wasn’t ready for
- Microsoft just took sides in AI policy fight
- OpenAI just disclosed something genuinely alarming
Security usually arrives after the race
This week, President Donald Trump hosts Chinese President Xi Jinping in Washington. Ahead of the summit, Bessent said the countries discussed a formal AI dialogue, including a U.S. proposal to flag national security incidents, CNBC reported.
Arora thinks Washington is skipping a step. He said the U.S. should first agree on what it actually wants before coordinating with anyone else.
Rules tend to trail technology. The internet’s architects put connecting computers ahead of protecting them, The Washington Post found in 2015. The bill for building first and securing later runs about $244 billion a year, according to Gartner’s forecast for 2026.
Arora is betting AI follows the same path, with Palo Alto selling the locks. If he is right, AI safety will not be settled by a lab’s release calendar. It will be bought, one security contract at a time.
Related: Anthropic researcher resigns and his reason is a warning to us all
