There’s a shift happening right now that a lot of companies don’t seem fully prepared for. AI isn’t just answering questions anymore. It’s starting to act. These so-called “agents” can access systems, move data around, and make decisions on their own. That’s powerful, but it also opens the door to a whole new set of problems.
Palo Alto Networks clearly sees that coming. It plans to acquire Portkey, a company focused on building a control layer for AI activity. The goal is to put some order around what is quickly becoming a chaotic situation inside enterprise environments.
Let’s put it in plain terms. Businesses are rolling out AI agents across different teams, often without a unified way to track what those agents are doing. One might be pulling customer data, another might be interacting with internal tools, and yet another might be making decisions based on incomplete or incorrect inputs. Multiply that across a large company, and things can get messy fast.
Portkey’s tech is meant to sit in the middle of all that. Think of it as a checkpoint. Every AI action flows through it, gets inspected, and can be allowed or blocked based on rules. After the deal closes, that system will become part of Palo Alto’s broader AI security platform, Prisma AIRS.
The pitch is simple. If companies are going to trust AI with real work, they need visibility and control. Otherwise, they’re basically handing over access to systems without knowing what’s happening behind the scenes.
There’s also a reliability angle here. Palo Alto is talking about keeping AI workloads running nearly all the time, with failovers and routing to keep things stable. That matters if AI is doing more than just helping out and actually running parts of the business.
Cost is another piece people don’t talk about enough. AI usage can get expensive in a hurry. Portkey claims it can help keep that under control by limiting usage and caching results so companies aren’t paying for the same work over and over again.
What stands out to me is how fast this is all moving. Not long ago, companies were experimenting with AI on the side. Now they’re pushing it into production and letting it handle real tasks. The problem is, security hasn’t really caught up yet.
At some point, companies are going to realize they can’t just deploy AI and hope for the best. They’ll need systems to monitor it, restrict it, and maybe even audit it like an employee. And here’s the twist. The only thing that might be fast enough to keep AI in check could be more AI.
The deal is expected to close later in Palo Alto Networks’ fiscal 2026. Once that happens, expect a lot more focus on controlling AI behavior instead of just building it. Because at this point, the bigger risk isn’t whether AI works. It’s what happens when it works too well without oversight.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz