Taming AI's Wild West: Tigera Unveils Lynx to Secure Autonomous Agents

SAN JOSE, Calif. – June 17, 2026 – As enterprises race to deploy autonomous AI agents, a new and chaotic digital frontier is emerging within their own networks. These agents, designed to act independently on behalf of users, represent a paradigm shift from the predictable workloads of the past. They are non-deterministic, can access any tool or data source, and operate in complex chains of delegation, creating a security and governance nightmare that existing systems were never designed to handle. Today, Tigera, the company behind the widely adopted Calico Open Source project, has stepped in to impose order with the launch of Tigera Lynx, a unified control plane designed to tame this untamed frontier.

The New Security Blind Spot: Autonomous Agents

The rush to integrate AI has created a significant blind spot for corporate security. Autonomous agents, while powerful, introduce a class of risks that leave security, platform, and AI teams in a difficult position. Unlike traditional applications, these agents are unpredictable. They can read untrusted input from the web, chain together tools in novel ways, and make decisions that diverge from their original programming, a phenomenon known as 'goal misalignment.'

This creates a perfect storm of vulnerabilities. Risks like prompt injection can turn a helpful agent into an internal saboteur, leading to data exfiltration or unauthorized actions. The rise of "shadow AI," where employees deploy agents without oversight, means many organizations have active, unmonitored autonomous entities with access to sensitive systems. One CISO at a financial services firm recently lamented the situation, noting, "It’s like giving a new intern the keys to the entire kingdom and hoping for the best. We have no visibility, no control, and no way to prove compliance."

This gap is precisely what Tigera aims to close. Lynx is engineered to give enterprises a single, unified platform to discover every agent in their Kubernetes estate, assess its security posture, assign it a secure sandbox, and enforce policy on every single action it takes—all without requiring developers to change a single line of agent code.

A Unified Control Plane for a Decentralized Workforce

Lynx functions as a central nervous system for an enterprise's AI agent ecosystem. It sits in the path of every call an agent makes—whether to another agent, an internal tool, or an external LLM—to authenticate, authorize, and audit the interaction before it executes. This is achieved through five core capabilities rolled into one control plane:

• Discovery and Observability: Using eBPF-powered auto-discovery, Lynx continuously scans the environment to find and catalog every agent, flagging and quarantining unregistered "shadow agents." It provides end-to-end observability, allowing teams to reconstruct an agent's actions through detailed OpenTelemetry traces.
• Posture Management: An AI-focused Cloud Security Posture Management (AI-CSPM) feature continuously evaluates each agent against security baselines, alerting teams to configuration drift or excessive permissions. It even includes a "Red Team Agent" that constantly probes for weaknesses.
• Verifiable Identity: Moving beyond shared secrets and long-lived API keys, Lynx integrates with enterprise identity providers like EntraID and Okta or uses SPIFFE/SPIRE to assign each agent a verifiable, short-lived cryptographic identity that is automatically rotated.
• Policy Enforcement: Using the Cedar policy language, administrators can define granular, default-deny policies that govern what agents can do. These rules are enforced at a gateway level, meaning misbehaving agents can be instantly quarantined or high-stakes actions can be flagged for human review.
• Anomalous Behavior Detection: By monitoring system activity at the kernel level, Lynx can detect suspicious behavior like credential theft or lateral movement, even if the agent possesses a valid credential.

The most critical feature for accelerating adoption is that this entire security and governance framework is applied externally. By not requiring code changes, Lynx removes a major point of friction between fast-moving AI development teams and the platform and security teams responsible for maintaining stability and control.

From Kubernetes Roots to AI's Future

Tigera is not a newcomer to securing complex, distributed systems. The company has spent over a decade building and maintaining Calico, the open-source project that has become a de facto standard for network security and observability in the Kubernetes world. This deep expertise in securing the infrastructure that powers modern cloud applications provides a powerful foundation for tackling the next generation of workloads: AI agents.

"For over a decade, Tigera's Calico platform has served Global 2000 companies running the largest Kubernetes platforms in the world, securing tens of millions of mission-critical transactions every day," said Ratan Tipirneni, CEO of Tigera. "AI agents are the next class of workloads: autonomous, distributed, and increasingly embedded in critical business processes. Lynx brings that same unified control and security rigor to AI agents. We're building on our core competency — securing mission-critical workloads at scale on Kubernetes, in a highly performant way."

This strategic evolution is a natural one. If Kubernetes is the decentralized power grid for modern enterprise applications, then AI agents are the autonomous devices plugging into it. Securing the grid itself is no longer enough; you must also be able to govern the behavior of everything connected to it.

Kernel-Deep Control: The eBPF Advantage

The technological heart of Lynx's "no code change" approach is its sophisticated use of eBPF (Extended Berkeley Packet Filter) and LSM (Linux Security Modules). These Linux kernel technologies allow Lynx to act as a secure, tamper-proof observer and enforcer at the most fundamental layer of the operating system. Instead of relying on application-level instrumentation, which can be bypassed or manipulated, Lynx watches every system call, network packet, and file access directly from within the kernel.

This provides an unforgeable source of truth for auditing and anomaly detection. As Tigera's Chief Technology Officer, Peter Kelly, explained, "Control only matters if it's enforced uniformly. Lynx gives every agent a cryptographic identity, scopes credentials to a single hop, and evaluates every LLM, MCP, and tool call against a default-deny policy at the gateway. Because we watch behavior with eBPF and LSM at the kernel, we can detect an agent going wrong even when it carries a valid credential, and produce a reproducible audit trail to prove it."

While eBPF provides unparalleled visibility into the consequences of an agent's actions—such as an unusual network connection or file access following a prompt injection—it offers a powerful layer of runtime defense that complements application-level security guardrails. This kernel-deep control ensures that even if an agent's logic is compromised, its blast radius is contained.

Paving the Way for Regulated AI Adoption

For enterprises in highly regulated industries like finance, healthcare, and government, the governance capabilities of Lynx are not just a feature but a prerequisite for adoption. The platform's ability to provide a forensic audit trail and enforce strict policies is critical for meeting compliance mandates under frameworks like GDPR, HIPAA, and SOC 2.

The fact that Lynx is already deployed in production at several top global banks underscores the market's urgent need for such a solution. These institutions, which cannot afford ambiguity in security or compliance, are leveraging the platform to build a defensible and auditable AI strategy. By providing the tools to manage identity, enforce data handling policies, and prove control over autonomous systems, Lynx is transforming AI agents from a high-risk experiment into a trusted, scalable component of the enterprise technology stack.