Cicada: Sub-ms Go Zero-Trust gateway and gVisor firewall for AI agents Apache 2 (github.com)

Cicada, an experimental project developed over eight months, introduces a sub-millisecond, zero-trust gateway and gVisor firewall specifically designed for autonomous AI agents. This proof-of-concept addresses critical vulnerabilities as autonomous AI systems often lack essential network-level protections, risking catastrophic failure loops and unauthorized data access. Unlike probabilistic content-layer filtering, Cicada employs deterministic network, cryptographic, and kernel-level controls, ensuring that all agent tool calls and requests pass through a rigorous enforcement framework with 15 sequential gates. Significantly, Cicada's Matthews Enforcement Framework (MEF) ensures enhanced security for AI operations, achieving performance improvements, including a drastic reduction in median latency from 265 ms to just 6.24 ms when optimized. This framework directly addresses prominent threats identified by OWASP for Autonomous Agents, implementing advanced measures such as recursive deobfuscation to prevent prompt injection attacks and a real-time quarantine system that detects and mitigates breaches within microseconds. The implications for the AI/ML community are profound, as Cicada provides a scalable architecture that other developers can adopt and adapt, ultimately fostering safer autonomous AI applications.