bytedance/vArmor

What it solves

vArmor is a cloud-native container hardening system designed to strengthen container isolation and reduce the kernel attack surface. It prevents container escapes, lateral movement, and unauthorized network access, which is particularly useful for protecting AI Agents and LLM-based applications from prompt injection-induced tool abuse or data exfiltration.

How it works

It implements security enforcers using a combination of Linux kernel technologies and network proxies. It leverages AppArmor, BPF LSM, and Seccomp for kernel-level mandatory access control (file access, process execution, and syscalls) and uses an Envoy-based sidecar proxy to intercept and control network egress traffic at L4 (TCP) and L7 (HTTP/HTTPS) levels, including TLS MITM for decrypted inspection.

Who it’s for

Cloud-native developers and security engineers managing Kubernetes clusters, especially those in multi-tenant environments or those deploying high-risk AI Agent workloads that require precise outbound network control.

Highlights

• Multi-layered Enforcers: Combines kernel-level (AppArmor, BPF, Seccomp) and network-level (Envoy) protection.
• Kubernetes Native: Follows the Operator design pattern and uses CRD APIs for workload hardening.
• AI Agent Protection: Specifically mitigates risks like prompt injection and unauthorized API calls for LLM apps.
• Advanced Network Control: Supports TLS MITM, HTTP header injection, and anti-Domain-Fronting protection.
• Flexible Security Models: Supports both Allow-by-Default and Deny-by-Default models with built-in rules.