LightGuard-Agent is an enterprise runtime firewall engineered to defend tool-calling AI agents, MCP servers, and RAG pipelines against:
- Indirect Prompt Injection (IPI)
- Steganographic & Unicode Homoglyph Evasion
- Delimiter & System Prompt Escapes
- Canary & Credential Exfiltration
- Multi-Turn Crescendo Jailbreaks
- Unauthorized Tool Escalation
| Version | Supported | Status |
|---|---|---|
0.7.x |
Yes | Current Stable Release |
0.3.x |
Yes | Research Benchmark Baseline |
< 0.3.0 |
No | Deprecated |
We treat all vulnerability reports with highest urgency. If you discover an evasion technique, false negative bypass, or code execution defect:
- Direct Communication: Send an email to
seotarek@gmail.comwith the subject[SECURITY] LightGuard Vulnerability Report. - Payload & PoC:
- Provide the exact adversarial payload or reproducible script.
- Specify the active configuration (thresholds, enabled tiers).
- Expected verdict vs. observed verdict.
- Disclosure Timeline:
- We will respond within 48 hours with an initial triage.
- A fix will be developed and verified against our automated benchmark suite before a public patch release is issued.
- We coordinate public CVE/GHSA disclosures responsibly.