## https://sploitus.com/exploit?id=34144148-D622-543F-BE67-DF27A78033A0
# Security Advisories
Public write-ups and PoCs for CVEs I've discovered through vulnerability research on AI/LLM infrastructure and open-source software.
All findings were responsibly disclosed through coordinated disclosure (GHSA / MITRE / huntr).
---
## Published CVEs
| CVE | Project | Type | CVSS | NVD |
|-----|---------|------|------|-----|
| [CVE-2026-32247](./CVE-2026-32247/) | graphiti-core (getzep) | Cypher Injection | 8.1 | [NVD](https://nvd.nist.gov/vuln/detail/CVE-2026-32247) |
| [CVE-2026-33980](./CVE-2026-33980/) | adx-mcp-server | KQL Injection | 8.8 | [NVD](https://nvd.nist.gov/vuln/detail/CVE-2026-33980) |
| [CVE-2026-34940](./CVE-2026-34940/) | KubeAI (substratusai) | OS Command Injection | 8.7 | [NVD](https://nvd.nist.gov/vuln/detail/CVE-2026-34940) |
| [CVE-2026-34975](./CVE-2026-34975/) | Plunk (useplunk) | CRLF Email Header Injection | 8.5 | [NVD](https://nvd.nist.gov/vuln/detail/CVE-2026-34975) |
All four are NVD-published. Full write-ups, root cause analysis, and standalone PoC scripts in each folder.
---
## Research context
These findings come from a broader effort to systematically audit agentic AI systems and MCP-based infrastructure. The vulnerability patterns found here (query language injection via MCP tool parameters, shell injection in Kubernetes operators) reflect systemic issues in the AI tooling ecosystem.
โ [All advisories I'm credited on](https://github.com/advisories?query=credit%3Aromain-deperne)