Palo Alto Networks Acquires Portkey to Build Unified AI Agent Security Gateway

--- headline: "Palo Alto Networks Acquires Portkey to Build Unified AI Agent Security Gateway" slug: palo-alto-portkey-ai-agent-security category: business story_number: "02" date: 2026-05-04 ---

Palo Alto Networks is betting that the next great cybersecurity battleground is not the firewall or the endpoint -- it is the AI agent. On April 30, the $130 billion cybersecurity giant announced its intent to acquire Portkey, the AI gateway startup already processing trillions of tokens per month, in a deal that industry observers at The New Stack have estimated in the $700 million class. Financial terms were not officially disclosed, but the strategic ambition is unmistakable: Palo Alto Networks wants to own the central nervous system through which every enterprise AI agent communicates, decides, and acts.

The Problem: Agents as Privileged Insiders

The acquisition targets what Palo Alto Networks sees as an existential gap in enterprise security. As organizations move beyond chatbots and copilots toward fully autonomous AI agents, these systems are increasingly operating as highly privileged insiders -- accessing sensitive data, executing complex workflows, and making real-time decisions across internal and external systems with minimal human oversight.

The attack surface is enormous and largely invisible. Fragmented security tools have forced enterprises to choose between innovation speed and safety, and most have chosen speed. Palo Alto Networks chairman and CEO Nikesh Arora framed the stakes bluntly in a LinkedIn post accompanying the announcement: "AI agents have become privileged insiders, reasoning and executing on behalf of users and companies. With that power comes a new category of risk. You cannot build an agentic enterprise without a centralized control plane to secure it."

Lee Klarich, Chief Product and Technology Officer of Palo Alto Networks, echoed the urgency in the official press release: "As autonomous agents join the enterprise workforce, they also become a new, unmanaged attack surface. By integrating Portkey into Prisma AIRS, organizations will be able to confidently deploy and govern AI agents. With Portkey, we are providing enterprises with visibility into all their agentic traffic, and enabling them to control and protect against agentic threats."

What Portkey Brings to the Table

Founded in 2023 by Ayush Garg and Rohit Agarwal, Portkey built an AI gateway that acts as a single access layer for more than 1,600 AI models and over 3,000 large language models and MCP tools via a unified interface. The platform routes requests across different models, manages traffic with automated failovers to achieve 99.99 percent uptime, and provides the kind of deep technical telemetry and audit logs that enterprise security teams demand.

The startup raised $15 million in its Series A funding round led by Elevation Capital just months before the acquisition announcement -- a sign of how quickly the deal came together. Portkey had evolved from a developer tool into core infrastructure, processing billions of API calls in production environments.

Rohit Agarwal, CEO and co-founder of Portkey, described the trajectory in the press release: "Scaling AI in production requires a delicate balance between total flexibility for developers and absolute control for security teams. By joining Palo Alto Networks, we will establish the AI Gateway as the foundational layer of the secure AI enterprise. Together, we will provide the infrastructure that allows every organization to deploy autonomous agents with the confidence that their data and operations are fully protected."

The Strategic Architecture

Portkey will be integrated into Prisma AIRS, Palo Alto Networks' AI runtime security platform, serving as the AI Gateway -- the choke point through which all agentic traffic flows. The combined platform is designed to inspect AI traffic and enforce security and governance policies at runtime, apply least-privilege identity controls to every agent interaction, provide centralized artifact management for versioning AI models and MCP servers, and eliminate cost overruns through caching and granular quotas.

The architecture effectively transforms the AI gateway from developer plumbing into an enterprise security checkpoint. Rather than bolting security onto agents after deployment, the approach embeds governance into the infrastructure layer itself.

Market Context and Competitive Implications

The deal lands amid a surge of activity in the agentic AI security space. Palo Alto Networks has also been linked to an acquisition of Armadin, signaling a broader strategy to build a comprehensive AI security stack. The company, which reported being trusted by more than 70,000 customers, is racing against CrowdStrike, Zscaler, and a growing cohort of startups to define the security architecture for autonomous AI systems.

The timing is deliberate. Enterprise spending on agentic AI infrastructure surged past $42 billion in Q2 2026 funding rounds alone, according to recent industry tallies, and companies deploying autonomous agents are discovering that traditional security tools are fundamentally inadequate for systems that reason, plan, and execute independently.

What to Watch Next

The transaction is expected to close in Palo Alto Networks' fiscal Q4 2026, subject to customary closing conditions. Following the close, Palo Alto Networks has committed to continuing support for existing Portkey customers while integrating the platform more tightly into Prisma AIRS.

The critical question is whether enterprises will consolidate their AI security stack around a single gateway or continue with the fragmented approach that Palo Alto Networks is betting against. If the Portkey integration delivers on its promise of security without latency penalties -- a non-negotiable requirement for agent-to-agent communication -- it could establish the AI gateway as the defining infrastructure layer of the agentic enterprise era. Watch for early adoption metrics when Palo Alto Networks reports its next quarterly earnings, and for competitive responses from the major cloud providers who may not want a cybersecurity vendor controlling the chokepoint of their customers' AI traffic.