Palo Alto buys Portkey

AI security is starting to look less like antivirus and more like traffic control. Once companies let autonomous agents call models, hit internal apps, and make decisions on their own, the real problem becomes supervision — who called what, with which data, under which policy, and whether anything drifted out of bounds. That is the gap Palo Alto Networks is trying to close with its April 30 deal to acquire Portkey and plug its AI gateway into Prisma AIRS. (paloaltonetworks.com) ### What is Portkey actually selling? Portkey sits between an application or agent and the models or tools that agent wants to use. Basically, it is a control plane for AI traffic: routing requests, enforcing policies, logging prompts and responses, handling reliability, and giving developers one layer (paloaltonetworks.com)RS. (paloaltonetworks.com) ### Why does an “AI gateway” matter now? Because enterprise AI has changed shape. A chatbot is mostly a user asking a model for text. An agent is different — it can chain steps, call tools, access data stores, and act with a non-human identity. That creates a new class of traffic that security teams wan(paloaltonetworks.com)es the checkpoint for that agent traffic. (paloaltonetworks.com) ### What changed with this deal? Palo Alto already had Prisma AIRS, its AI security platform. Portkey gives it a more mature gateway layer to sit in front of models and agents. The company says that, once the deal closes, Portkey’s full gateway will be integrated into Prisma AIRS so customers can discover AI use, apply guardrails, watch behavior, and manage agents at scale from one place. (paloaltonetworks.com) ### How big is Portkey’s footprint? Big enough to make this more than a tiny feature pickup. Palo Alto says Portkey already processes more than 2 billion API calls every day. It also says the platform supports more than 160 models and providers. That matters because the hard part for enterprises is not picking one model — it is governing a messy multi-model stack without rebuilding controls every time teams switch vendors. (paloaltonetworks.com) ### Why is Palo Alto making so many AI moves? Because it is trying to turn “AI security” into a full platform category before rivals do. The Portkey announcement lands after Prisma AIRS 3.0 in March, the completed CyberArk acquisition in February, the Chronosphere acquisition in January, and Palo Alto’s April completion of Koi. Put together, the pattern is pretty clear: identity, observability, endpoint, and now gateway control for agents. (paloaltonetworks.com) ### So what does customers’ setup look like? In Palo Alto’s version, an employee or software agent sends a request, the gateway checks policy, routes the call to the right model, watches inputs and outputs, and logs what happened for governance and debugging. Think of it like an air-traffic tower for model calls — not flying the plane, but deciding what can take off, wher(paloaltonetworks.com)ary. That is an analogy, but it maps pretty closely to the product language here. (paloaltonetworks.com) ### What’s the catch? An AI gateway only helps if companies actually force traffic through it. Shadow AI, direct API keys, and agent sprawl can still bypass central controls. And Palo Alto did not disclose the price, so the market still does not know how much it paid for the capability. But strategically, the logic is straightforward — if agents become normal enterprise workers, the checkpoint in front of them becomes valuable fast. (paloaltonetworks.com) ### Bottom line? This is Palo Alto betting that the next important security layer is not just around users or devices, but around machine-driven decisions. Portkey gives it a way to treat AI-agent activity as something visible, governable, and billable — which is exactly where enterprise security spending may be heading. (paloaltonetworks.com)