SoylentNews is people
SoylentNews
https://www.theregister.com/2026/02/18/palo_alto_q2_26/
If enterprises are implementing AI, they're not showing it to Palo Alto Networks CEO Nikesh Arora, who on Tuesday said business adoption of the tech lags consumer take-up by at least a couple of years – except for coding assistants.
"Consumers are far outstripping enterprise for the moment, but we expect enterprise will surely and slowly get on that bandwagon," he said on the company's Q2 earnings call.
Arora likened business uptake of AI to the cloud computing shift, which he said took two or three years before enterprises started migrating applications.
"Right now ... tell me how many enterprise AI apps are you using which are driving tremendous amounts of throughput," he asked, and answered himself "I can't think of anything but coding apps."
Coding apps aren't great for Palo Alto's business because they don't generate a lot of network traffic to which it can apply its security smarts. Arora thinks his security vendor peers know this.
"We're all laying the groundwork right now. It is ... sort of an arms race to try and see who can get the AI security sort of platform up and running as quickly as we can."
But the limited enterprise AI adoption Arora has seen does pose some immediate challenges to Palo Alto.
"There is now enterprise adoption that we're beginning to see where customers are running perhaps millions of tokens in one or two particular applications they're working with some of the LLM providers on, and that's where we see the traffic," he said. That traffic is on the LAN and the CEO doesn't think existing networks struggle to handle it.
"I think the challenge right now is consolidating that traffic," he said. "How do you get all the AI traffic to be in one place? So you can understand it, provide visibility, look at the ability to control it and be able to act on it."
The CEO said that as this sort of AI-related traffic grows "it needs a different set of controls and tools."
Palo Alto is already getting its hands on those tools, as on Tuesday put to bed rumours it would acquire agentic AI endpoint security startup Koi by announcing it's done the deal.
Arora pointed to Palo Alto's recent acquisitions of Chronosphere and CyberArk as further evidence of the company's moves to ensure it builds a portfolio of products to secure the AI enterprises will eventually implement.
The CEO expressed confidence Palo Alto has the products it needs today, saying customers know they can't prepare for AI if they are running a tangle of security tools and are therefore consolidating to the kind of platforms the company offers.
Demand for those products helped Palo Alto to $2.6 billion Q2 revenue for the quarter, which represented 15 percent year-over-year growth.
Execs pointed to the success of the company's subscription offerings, noting 23 percent growth in remaining performance obligations, which now stand at $16 billion. And they predicted Q3 revenue would grow at least 28 percent to land between $2.941 billion and $2.945 billion.
All of those nice numbers didn't impress investors, who knocked six percent off the company's share price – perhaps because they weren't thrilled by predictions that profits will ease.
(Score: 4, Interesting) by VLM on Saturday February 21 2026, @04:47PM
This is one of those articles that was dumbed down by the journalism filter and/or the AI filter making it hard to understand than if it was not dumbed down.
However, given the above, I "think" that the article is talking about companies that do dumb things like install a MCP interface on an internal server then expose it to the entire internet or too much of the internet to do AI slop stuff.
If you're not already involved, MCP is analogous to a REST API for AI LLMs to connect to. So rather than cut and paste your entire database schema attached to every prompt you connect an LLM once to a database server using an MCP interface. In theory the people doing this will be intelligent and security conscious, in practice we know how this turns out every time for the past couple decades. So instead of permitting read only access to one low security database's table schemas to one IP address, they'll give full read-write access to all databases including payroll and CC number storage and open it to the entire internet "because it worked, and it was too much work to secure".
And yeah unless you're doing something really expensive or really weird or both, MCP traffic is pretty light and intermittent.
The kind of AI slop you can do with a MCP server, on a DB anyway, would be "hey AI give me the exact SQL statement to join our illegal, forbidden by PCI-DSS table of unencrypted customer credit card numbers to our unsecured and unaudited list of customer orders that haven't been marked as billing process initiated" or something like that.
The kind of people who need AI slop to do basic coding are going to correlate strongly with the kind of people who can't secure their way out of a wet paper bag with a flashlight and map or something like that. Expect an absolute flood of "cybersecurity" stories relating to MCP, its going to be very fertile ground assuming the bubble lasts long enough.