LLMs aren’t accurate enough to displace key segments such as security operations—and simply can’t leverage the proprietary data of a vendor like Palo Alto Networks, CEO Nikesh Arora said Tuesday.
Investor fears that AI poses more of a risk than an opportunity for cybersecurity vendors are unfounded, with LLMs unlikely to rival the capabilities of security products in the foreseeable future, Palo Alto Networks CEO Nikesh Arora said Tuesday.
Arora made the comments while discussing the widespread concerns from investors that AI could become a competitive threat for the software industry, which has led share prices for many top vendors to sink in recent weeks.
[Related: The 2026 Security 100]
During Palo Alto Networks’ quarterly call Tuesday, Arora told analysts that while GenAI and AI agents are already proving to be massively helpful for security products and teams, there are clear limitations on what LLMs can do.
“I’m still confused why the market is treating AI as a threat” to the cybersecurity industry, he said, while adding that he “can’t speak for all of software.”
For starters, LLMs aren’t accurate enough to fully replace key segments such as security operations, Arora said.
“Until [LLMs] get to 99.9-percent accuracy, they’re not a threat to delivering security,” he said.
Meanwhile, many security tools—including from Palo Alto Networks and its broad platform—have a major leg up through having access to real-world customer data for training its AI models, Arora noted.
Vendors such as Palo Alto Networks are utilizing “domain-specific data, based on threats we see out in the environment,” and then using the data to determine how to best protect the customer, he said.
“To the extent we are creating proprietary data in security, that is not going to be replaced by an LLM,” Arora said.
Ultimately, while “every security product has some version of a Copilot that now runs in tandem with the product,” that doesn’t mean that major disruption to the cybersecurity industry is on the horizon due to the rise of AI, according to Arora.
“I don’t think it’s going to replace the security product anytime soon,” he said.
AI Security Push
In addition to expanding its AI-powered security capabilities, Palo Alto Networks is also continuing to double down on providing security for the usage of AI and agentic tools.
Earlier Tuesday, Palo Alto Networks announced a deal to acquire Koi, a startup offering capabilities for boosting visibility and protection for AI agent usage on endpoint devices.
The acquisition—the terms of which were not disclosed—will give Palo Alto Networks technology in the new cybersecurity category of “agentic endpoint security,” the company said in a news release. The cybersecurity giant plans to integrate Koi’s capabilities into its AI security platform, Prisma AIRS, the vendor said.
During the company’s call Tuesday, Arora said that Prisma AIRS is already “rapidly scaling” with more than 100 customers as of the end of its fiscal second quarter, which ended Jan. 31.
“This is much faster than we [scaled] in cloud security,” he said.
For the company’s fiscal Q2 overall, Palo Alto Networks reported that revenue climbed 15 percent, from the same period a year earlier, to reach $2.59 billion. That came in just above the Wall Street analyst consensus estimate for the quarter.
Earnings during the quarter, meanwhile, reached $1.03 per share, beating estimates by 9 cents per share.
CyberArk Acquisition
The financial report followed the completion last week of Palo Alto Networks’ $25 billion acquisition of identity security powerhouse CyberArk.
The acquisition, which is by far the largest in the vendor’s history, is pivotal both for enabling Palo Alto Networks to expand into identity security and for boosting the company’s platform for securing agents, Arora said Tuesday.
“We bought CyberArk because when AI agents start logging in at machine speed, logging in becomes the primary attack vector,” he said. “We believe we are now the only company that can verify the ‘who’ and secure the ‘what’ simultaneously.”
