RamLab

AIShield Runtime

Prototype 2025 PHP, Python, MySQL

Behavioral firewall for AI agents with prompt/tool policy and anomaly detection.

Experiment Log

Aim

Detect and block risky agent behaviors in real time.

Hypothesis

Profiles + rules stop most attack classes.

Method

Rule engine then ML anomalies.

Results

Caught simulated exfil attempts in sandbox.

Bill of Materials

  • PHP
  • Python
  • MySQL

Lab Notes

Add shadow agent alerts.