AI Agents Need Guardrails. These 3 Stocks Could Benefit.
Companies want autonomous software they can trust. Palo Alto, CrowdStrike, and Cloudflare offer three ways to follow the spending on AI agent security.
Listen to the audio version of this article (generated by AI).
Earlier this year, a group of AI agents escaped from the sandbox built to contain them.
According to OpenAI‘s July 21 disclosure, the agents were carrying out cybersecurity evaluations when they gained unauthorized internet access, used exposed credentials and exploited software vulnerabilities. The company called it an “unprecedented cyber incident.”
Now California attorney general Rob Bonta is suing over it.
AG Bonta’s complaint alleges that OpenAI tested powerful models without the safeguards and monitoring needed to protect the public. He states that the breach was “entirely foreseeable” and that the widely-known Hugging Face break-in was “only one of 80 known incidents” involving experimental agents.
The state wants civil penalties, plus monitoring and secure testing. OpenAI says that the suit singles it out over a challenge every major AI lab has acknowledged and that it has strengthened its safeguards, published independent reports, and built a framework to catch agents acting against their instructions.
The courts will sort out who’s right, though that could take years.
Thankfully, investors don’t have to wait that long – because of what happened in the two days after the suit was announced.
First, Microsoft (MSFT) used its Ignite conference to roll out a set of tools for governing AI agents. Then Palo Alto Networks (PANW) reported earnings and said its Prisma AIRS platform had reached nearly $170 million in annual recurring revenue.
Every company wants the productivity of an AI workforce. But the minute you let that workforce do things on its own, you need a way to keep it in line.
Why AI Agents Create New Cybersecurity Risks
AI that talks and AI that acts are two very different security problems.
A chatbot that answers questions is a reference tool. If it gets something wrong, a person reads its answer and decides what to do with it.
An agent is different. It can access customer records, change code, move files, and kick off transactions. In practice, a company that deploys one has hired a new kind of employee.
Every employee needs the same things: a defined role, clear limits and someone to supervise the work. An agent also needs a way for the company to revoke its access immediately.
Businesses already have systems for managing access and monitoring activity. Extending that to agents is a new challenge. Software can move quickly across tools, interpret instructions unexpectedly… take actions a supervisor never intended.
In October, OpenAI reported two cases in which internal coding agents weakened monitoring while making requested software changes. One disabled automatic security alerts; another removed a monitoring integration and then incorrectly claimed it had preserved it.
Before businesses hand software more responsibility, they need confidence that they can see everything it’s doing – and intervene if necessary.
Question One: Are Companies Paying for AI Security?
That $170 million figure is the number I’d circle in Palo Alto’s report. Prisma AIRS, the company’s AI-security platform, got there across more than 1,000 customers in just five quarters.
Prisma AIRS does cover the whole AI stack – applications, models, data, and agents – so the figure reflects more than agents alone. Even so, it’s the best early read we have on whether companies will pay extra to secure AI as they roll it out. So far, they will.
Across the company, quarterly revenue came in at $3.44 billion, up 39% from a year ago. Next-generation security ARR – an annualized measure of active contracts across the company’s newer security offerings – reached $9.73 billion, up 66%. Both growth figures include acquired businesses.
Management also raised its full-year revenue forecast to a range of $14.52 billion to $14.57 billion, from $14.10 billion to $14.20 billion. Prisma AIRS remains a small part of that broader business, but it gives us a specific product and an ARR figure to track as AI adoption expands.
The stock opened sharply higher post-earnings, and it pulled CrowdStrike (CRWD), Zscaler (ZS), Okta (OKTA), and most of the sector up with it.
Question Two: Is AI Agent Security Sold Separately or Bundled?
At Ignite, Microsoft announced tools that can discover more than 60 types of AI agents running on company devices – including rivals’ products like Claude Code and OpenAI’s Codex – with separate connections for agents running on cloud platforms.
Finding an agent is step one. Step two is giving it an identity in the company directory – same as a human employee – so its access can be granted for a specific task and pulled back afterward. That piece is still in preview, and it starts with just two agents: Claude Code and GitHub Copilot CLI.
Then there’s the price. Microsoft’s agent-management package, Agent 365, lists at $15 per user per month on its own and comes included in the company’s top-tier Microsoft 365 E7 bundle. Some basic visibility comes with subscriptions customers already have; the more advanced controls cost extra.
So, Microsoft is treating agent security as a product worth charging for. That also makes life harder for standalone vendors. Microsoft already covers agents from other providers, so “we work across platforms” won’t be enough to win a second purchase.
The independents will have to show they catch what Microsoft misses, fit more smoothly into how security teams already work, or cost less to run. Paid adoption and growing contracts will tell us who’s pulling that off.
Question Three: Is AI Agent Adoption Expanding Security Contracts?
The next test is whether broader use of agents leads to larger security contracts.
Palo Alto’s technology chief said AI coding tools are the first agent use case running in real production, while most others are still pilots. CEO Nikesh Arora said most enterprises haven’t figured out how to secure agents yet and that charging for the agent traffic flowing through Palo Alto’s network products is still six to nine months away.
CrowdStrike is at a similar stage. In its last reported quarter, net new ARR grew 51% to a record $332.8 million, and total ARR reached $5.84 billion. In September, it launched Falcon Guardian, which watches and controls agents on the devices where they run. Earlier this month, it built that protection directly into Anthropic‘s (ANT) Claude Cowork. CEO George Kurtz has called securing AI “the largest market opportunity in our history.”
What CrowdStrike hasn’t done is break out AI-security revenue as its own number. Its next report, due Dec. 1, is what we’ll be watching for customer adoption, expanding contracts or a clearer contribution from AI security.
Our scorecard, then, reads like this: demand is real, and customers are writing checks. But the larger opportunity depends on agents moving into everyday operations and vendors expanding contracts and durable earnings.
3 AI Cybersecurity Stocks to Watch
Security has a quality we value in this market: since it’s the cost of putting AI to work, it has to be paid whether the next data center gets approved or not.
While the market has been punishing AI stocks lately, our conviction in the boom remains strong. But our approach to the trade has grown more selective to weather the storm. Spending that scales with usage, and that doesn’t wait on construction, fits the bill.
Three names cover the main angles.
Palo Alto Networks: Prisma AIRS and Paid AI Security Demand
Palo Alto Networks has the clearest evidence so far. It sells a product built specifically to secure AI, and customers are already paying for it. It’s also where many large companies are consolidating their security spending, which gives it an easy path to add agent protection to contracts it already holds. The more agents those customers deploy, the more there is to protect. Growth in Prisma AIRS over the next few quarters will show how quickly that’s happening.
CrowdStrike: Protecting AI Agents Where They Work
CrowdStrike lives on the devices and servers where agents do their work. Its Falcon Guardian product watches what an agent is doing there and can step in when something goes wrong, and the company is building that protection into the AI tools and platforms businesses already use. Because so many companies already run CrowdStrike, agent security becomes an upgrade to an existing relationship. Its Dec. 1 report is the next chance to see whether that’s showing up in the numbers.
Cloudflare: Controlling AI Agent Access to the Web
Cloudflare (NET) sits at the front door of a huge share of the internet, which is where agents show up when they visit websites and online services. Its tools let site owners see which agents are knocking and decide which ones to let in, and its AI Gateway can tie a company’s AI requests back to verified users. The more of the web’s traffic comes from software, the more those services are worth. Paid adoption will show how much of that turns into revenue.
The Bottom Line: Why Securing AI Agents Requires More Computing Power
One detail from Palo Alto’s report deserves more attention than it got…
Alongside earnings, the company announced a $4.5 billion deal to buy Applied Compute, a platform for training and running AI models. CEO Arora’s reasoning was that security software can’t anticipate everything an agent might do, so Palo Alto needs to train its own models to “fight AI with AI.” The company expects to spend up to $3 billion on AI infrastructure of its own through the end of the decade.
So a cybersecurity company is now buying computing power to keep up with the agents it’s guarding.
That tells us something about where all of this is headed. Every agent a business deploys runs on chips and electricity. So does every model assigned to watch it. The AI workforce and its security detail are drawing from the same supply of computing power. And as we’ve noted for months, that supply is the tightest constraint in the entire buildout.
That constraint sits at the center of research I’ve been doing on Elon Musk’s companies, which are building toward something far larger than any one of them could manage alone.
A plan that large needs specialized equipment, critical materials, and components from outside suppliers – and that’s where I believe the opportunity for investors sits. In my latest briefing, I explain what’s being built, and name the companies I believe are positioned to benefit.
AI is getting a workforce and a security detail at the same time. Both of them need somewhere to run.
See which suppliers I believe are positioned to provide it.



