🛡️ Vulnérabilités
Explorateur du knowledge graph de Hermes
🌐 Carte topologique interactive
Section intitulée « 🌐 Carte topologique interactive »Utilisez le graphe de force interactif ci-dessous pour naviguer entre les entités interconnectées : vulnérabilités CVE, motifs d’attaque agentiques (AAP), études de recherche académiques, techniques MITRE ATT&CK et composants logiciels.
- Panoramique & zoom : glissez la souris pour vous déplacer, molette ou boutons pour zoomer.
- Recherche & filtrage : saisissez un identifiant pour surbriller les nœuds ou cliquez sur les catégories pour filtrer par type.
- Tiroir d’inspection : cliquez sur un nœud pour ouvrir le volet d’intelligence avec liens directs vers les dossiers, métadonnées et relations connectées.
- Isolation de voisinage : cliquez sur « Isoler le Voisinage » dans le panneau pour focaliser l’affichage sur les nœuds directement connectés.
⚡ API REST publique du graphe
Section intitulée « ⚡ API REST publique du graphe »Hermes Codex expose des points de terminaison JSON statiques et lisibles par machine, générés lors de la compilation :
- Graphe topologique complet :
/api/graph.json— nœuds, arêtes, degrés et typologie d’entités. - Métriques & statistiques :
/api/stats.json— décomptes d’entités, métriques relationnelles et taux de preuve.
# Interroger le Knowledge Graph d'Hermes via curlcurl -s https://hermes-codex.vercel.app/api/stats.json | jq .curl -s https://hermes-codex.vercel.app/api/graph.json | jq '.nodes[] | select(.type == "attack_pattern")'🔍 Analyse de voisinage phare : CVE-2026-59822
Section intitulée « 🔍 Analyse de voisinage phare : CVE-2026-59822 »Examinez le voisinage à 1 et 2 sauts centré sur la vulnérabilité LiteLLM (contournement d’authentification HTTP Streamable MCP), reliant les composants affectés, les techniques MITRE, les artefacts forensiques et les règles de détection Sigma :
CVE-2026-59822: LiteLLM MCP Streamable HTTP Auth BypassVULNÉRABILITÉ
Enterprise LLM proxy gateway supporting Model Context Protocol (MCP) streamable endpoints and unified LLM APIs.
🔍 Pourquoi cette relation ? (Preuves & Provenance)
“Directly confirmed by vendor advisory GHSA-59822 and federal advisory in CISA KEV.”
- [government_confirmation]CISA verified active exploitation in the wild and mandated federal remediation deadline in KEV entry. — Source : Cybersecurity & Infrastructure Security Agency (CISA): CISA Adds CVE-2026-59822 to Known Exploited Vulnerabilities Catalog (Fiabilité : VERY_HIGH)
- [vendor_confirmation]LiteLLM vendor advisory explicitly documents that any bearer token string was accepted on streamable endpoints. — Source : BerriAI LiteLLM Team: Security Advisory: Streamable HTTP Authentication Bypass in LiteLLM MCP Server (Fiabilité : VERY_HIGH)
Adversarial subversion of structured tool execution arguments (SQL, Shell, Filepath) passed from an LLM agent to host OS tools or MCP endpoints.
🔍 Pourquoi cette relation ? (Preuves & Provenance)
“Hijacking MCP endpoints enables attackers to supply crafted tool execution parameters.”
- [vendor_confirmation]LiteLLM vendor advisory explicitly documents that any bearer token string was accepted on streamable endpoints. — Source : BerriAI LiteLLM Team: Security Advisory: Streamable HTTP Authentication Bypass in LiteLLM MCP Server (Fiabilité : VERY_HIGH)
Adversaries search compromise victims for unsecured credentials in files, environment variables, or memory.
🔍 Pourquoi cette relation ? (Preuves & Provenance)
“Unauthenticated MCP access allows scraping upstream model API keys and internal environment variables.”
- [vendor_confirmation]LiteLLM vendor advisory explicitly documents that any bearer token string was accepted on streamable endpoints. — Source : BerriAI LiteLLM Team: Security Advisory: Streamable HTTP Authentication Bypass in LiteLLM MCP Server (Fiabilité : VERY_HIGH)
HTTP traffic logs demonstrating connections to /mcp/streamable with arbitrary Bearer tokens bypassing validation.
🔍 Pourquoi cette relation ? (Preuves & Provenance)
“Access logs record HTTP POST requests with missing or dummy Authorization headers.”
- [vendor_confirmation]LiteLLM vendor advisory explicitly documents that any bearer token string was accepted on streamable endpoints. — Source : BerriAI LiteLLM Team: Security Advisory: Streamable HTTP Authentication Bypass in LiteLLM MCP Server (Fiabilité : VERY_HIGH)
Detects anomalous streamable HTTP session initialization to LiteLLM endpoints with missing or dummy bearer tokens.
🔍 Pourquoi cette relation ? (Preuves & Provenance)
“Sigma rule SIG-MCP-042 flags unauthenticated session establishment requests.”
- [vendor_confirmation]LiteLLM vendor advisory explicitly documents that any bearer token string was accepted on streamable endpoints. — Source : BerriAI LiteLLM Team: Security Advisory: Streamable HTTP Authentication Bypass in LiteLLM MCP Server (Fiabilité : VERY_HIGH)
Attacker registers rogue MCP tools or skills with weaponized docstrings and deceptive metadata that trick the model into routing sensitive user tasks to attacker-controlled functions.
🔍 Pourquoi cette relation ? (Preuves & Provenance)
“Unauthenticated MCP access allows registering rogue tool definitions with weaponized descriptions.”
- [vendor_confirmation]LiteLLM vendor advisory explicitly documents that any bearer token string was accepted on streamable endpoints. — Source : BerriAI LiteLLM Team: Security Advisory: Streamable HTTP Authentication Bypass in LiteLLM MCP Server (Fiabilité : VERY_HIGH)
Exploitation of unauthenticated, unsigned inter-agent communication channels to forge delegation directives, impersonate orchestrator agents, and command worker subagents.
🔍 Pourquoi cette relation ? (Preuves & Provenance)
“Compromising the MCP streaming proxy allows injecting spoofed responses into peer agent message flows.”
- [vendor_confirmation]LiteLLM vendor advisory explicitly documents that any bearer token string was accepted on streamable endpoints. — Source : BerriAI LiteLLM Team: Security Advisory: Streamable HTTP Authentication Bypass in LiteLLM MCP Server (Fiabilité : VERY_HIGH)
Enterprise LLM proxy gateway supporting Model Context Protocol (MCP) streamable endpoints and unified LLM APIs.
🔍 Pourquoi cette relation ? (Preuves & Provenance)
“Confirmed security vulnerability in LiteLLM Proxy & MCP Server documented in Hermes dossier.”
- [vulnerability_report]
- [government_confirmation]CISA verified active exploitation in the wild and mandated federal remediation deadline in KEV entry. — Source : Cybersecurity & Infrastructure Security Agency (CISA): CISA Adds CVE-2026-59822 to Known Exploited Vulnerabilities Catalog (Fiabilité : VERY_HIGH)
🎯 Voisinage phare d’attaque agentique : AAP-003
Section intitulée « 🎯 Voisinage phare d’attaque agentique : AAP-003 »Explorez la topologie relationnelle centrée sur l’Altération et l’Injection de Paramètres d’Outils (AAP-003) :
AAP-003: Tool Parameter Tampering & Built-in BypassMODÈLE D'ATTAQUE
📚 Répertoire des entités
Section intitulée « 📚 Répertoire des entités »🎯 Motifs d'attaque
📐 Méthodologie formelle
🔬 Études fondamentales IA