Wardline: Auto-Blocking Proxy for Securing AI Agents
This open-source Go solution detects and blocks compromised AI agents in real time.
1What is Wardline?
Wardline is a proxy solution that monitors and enforces security policies for AI agent calls. It runs as a single static Go binary without requiring databases or additional services. It controls all calls made by AI agents to MCP servers, tools, and gRPC services. Developed as open-source, Wardline provides core security functions like authentication, policy enforcement, budget control, and auditing.
2Real-Time Anomaly Detection
Wardline detects anomalies using four statistical methods: rate spikes, novel tool usage, deny-rate spikes, and a combined ML score. These methods don't require training data or external models. When an anomaly is detected, the auto_block feature activates to temporarily block calls from the affected identity. This goes beyond alerts to actively prevent attacks.
3Multiple Policy Support
Wardline supports three policy engines: static YAML, embedded OPA/Rego, and embedded AWS Cedar. Users can choose their preferred policy with a single config key. All policy enforcement runs within the same binary without external processes or network hops. This flexibility makes it adaptable to different security needs.
4Identity and Access Management
Wardline provides identity management with short-lived RS256 JWT tokens, refresh tokens, and JWKS rotation. It offers features like OIDC and mTLS-SPIFFE bootstrapping, Kubernetes-style RBAC, SCIM 2.0 provisioning, and end-to-end tenant isolation. The X-Wardline-Identity header tracks each call by identity for policy enforcement.
5Two-Tier Rate Limiting
Wardline offers two-tier rate limiting: per-identity and per-tenant. Both limits must be satisfied for a call to proceed. This dual-layer approach controls resource usage for both individual agents and entire tenants. Rate limits help prevent overloading and potential attacks.
6Compliance and Auditing
Wardline maintains structured JSON audit logs. The wardline export-evidence command creates checksummed, RSA-signable evidence bundles for auditors. Configurable retention meets corporate compliance requirements. The wardline infer-policy feature can generate a basic allow-list from observed traffic patterns.
7Observability and Federation
Wardline enables cross-instance correlation through signed, anonymized anomaly summaries. It supports OpenTelemetry tracing, a live web dashboard, and HA multi-replica deployment with shared state over Postgres. The demo environment spins up quickly with make demo, showing anomaly blocks and policies in the integrated read-only dashboard.
8A New Standard in AI Security
Wardline provides a comprehensive solution for securing AI agents. With real-time anomaly detection, flexible policy engines, and robust identity management, it enhances the security of AI calls. Being open-source allows for community contributions and rapid development. For cloud-based AI applications especially, Wardline could become a critical security component.