Palo Alto buys Portkey for AI
- Palo Alto Networks said on April 30 it plans to buy Portkey, an AI gateway startup, and fold it into Prisma AIRS to govern AI agents. - The deal is set to close in Palo Alto’s fiscal Q4 2026. Portkey says its stack is already used by more than 3,000 GenAI teams. - This matters because Palo Alto is turning agent security into a control-plane problem — not just a model-safety or prompt-filtering problem.
AI security is starting to look a lot less like chatbot moderation and a lot more like identity and infrastructure. That is the real story behind Palo Alto Networks’ move to buy Portkey. Palo Alto announced the deal on April 30, 2026, and the point is pretty clear — if companies are going to let autonomous agents call tools, touch data, and trigger actions, they need a central place to see and control that traffic. Portkey gives Palo Alto a way to build exactly that inside Prisma AIRS. ### What is Portkey actually selling? Portkey is an AI gateway company. In plain English, that means it sits between an enterprise app or agent and the models, tools, and providers that app calls. It handles routing, observability, guardrails, governance, logging, and policy controls. Portkey pitches this as a production stack for GenAI builders, and on its site it says more than 3,000 teams use the platform. (paloaltonetworks.com) ### Why would Palo Alto want that? Because the hard part of enterprise AI is no longer just “which model should we use?” The hard part is “who is allowed to do what through the model?” Palo Alto’s pitch is that autonomous agents create a new unmanaged attack surface. If an agent can look up customer records, call a database, open a ticket, or trigger a workflow, the security problem starts to resemble IAM and runtime control — with attribution, permissions, and policy enforcement wrapped around every step. (portkey.ai) ### Why is the word “gateway” doing so much work? Because a gateway becomes the choke point. That is where you can inspect prompts and responses, log which agent called which tool, apply guardrails, cap usage, and block risky actions before they fan out across the rest of the stack. Think of it like an API gateway, but for model calls and agent workflows. The model is not the whole system anymore — the gateway is where governance becomes enforceable. (paloaltonetworks.com) ### How does this fit Prisma AIRS? Palo Alto says Portkey will become the AI Gateway inside Prisma AIRS, which is its AI runtime security platform. The company’s own language is revealing here — it calls the gateway a “mission-critical control plane for autonomous agents.” That framing matters. Palo Alto is not treating agents as a side feature inside existing app security. It is treating them as first-class actors that need centralized policy and visibility. (paloaltonetworks.com) ### Is this just about prompt injection? No — and that is the bigger shift. Prompt injection and data leakage still matter, but Palo Alto is broadening the problem. Once agents can chain tools and act on behalf of users or systems, security teams need to know what permissions the agent inherited, what action it took, and whether that action should have been allowed. That pushes AI security closer to identity governance, auditability, and delegated authorization. (paloaltonetworks.com) Portkey’s existing features — RBAC, API key management, audit logs, budgets, and request restrictions — line up neatly with that need. ### What else is Palo Alto signaling? That it wants to own more of the agent stack. Its 2026 press releases show a pattern — Prisma AIRS 3.0 in March, Koi for the “agentic endpoint” in February and April, and now Portkey for the gateway layer. Put together, Palo Alto is building security around where agents run, what they connect to, and how their actions get controlled. (portkey.ai) ### What is the catch? A control plane only helps if companies actually route agent traffic through it. Plenty of teams still build AI features in a messy, provider-by-provider way. So the acquisition does not solve agent risk by itself. But it does show where the market is heading — toward centralized enforcement points instead of scattered bolt-on filters. (paloaltonetworks.com) ### Bottom line? Palo Alto is betting that enterprise AI will be secured the same way serious infrastructure gets secured — with identity, policy, logging, and a hard control point in the middle. Portkey gives it that middle. (paloaltonetworks.com 1) (paloaltonetworks.com 2)