CVE-2026-35029: LiteLLM Gateway Authorization Bypass Enables RCE

Check Point Research disclosed multiple high-severity vulnerabilities in LiteLLM, a widely used open-source LLM gateway. The flaws allowed any authenticated user to modify the proxy configuration, extract credentials, and execute arbitrary code.

Christopher Ort
By Christopher Ort · 3 min read
Share
A security researcher examining server logs on multiple monitors in a dimly lit operations center
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. What you should do
  5. What it means
  6. Sources
i10X Quick Take

Quick Take

  1. What: Missing authorization check in LiteLLM gateway /config/update endpoint (CVE-2026-35029) and SSRF in Clientside LLM Credentials feature (CVE-2026-59823)
  2. Who is affected: Organizations running LiteLLM gateway versions before 1.83.9
  3. Severity: High-impact vulnerabilities enabling arbitrary file read, admin account takeover, and remote code execution
  4. Fixed version: LiteLLM 1.83.9
  5. What to do: Upgrade to LiteLLM version 1.83.9 or later immediately

In plain English

LiteLLM is a popular open-source tool that sits between applications and AI model providers like OpenAI or Anthropic, handling credentials and routing requests. Security researchers found serious flaws in version 1.77.7 that let any logged-in user, even those with limited permissions, change the system’s configuration. This could let attackers read sensitive files, take over administrator accounts, run malicious code on the server, or steal API keys for AI services. The developers fixed these problems in version 1.83.9.

What you should do: If you use LiteLLM gateway in your organization, you should upgrade immediately. If you only use AI chatbots as an end user, this does not affect you.

What happened

Check Point Research conducted a security assessment of LiteLLM version 1.77.7, an open-source LLM gateway that routes requests between applications and model providers. The researchers discovered multiple high-severity vulnerabilities tracked as CVE-2026-35029 and CVE-2026-59823.

The most critical flaw stemmed from a missing role-based access control check on the /config/update endpoint. This endpoint is designed to allow runtime modifications to the proxy configuration, including environment variables, model lists, and general proxy settings. While other administrative endpoints in LiteLLM explicitly verify that the authenticated user has the PROXY_ADMIN role, the /config/update endpoint relied only on generic authentication without checking the user’s role.

The vulnerability was compounded by an overly permissive route allowlist for UI sessions. When users log into the LiteLLM UI, the system generates a temporary API key associated with a special “litellm-dashboard team. The authorization logic classified tokens based on their team ID, and tokens classified as ui received a more permissive route check. The ui_routes list included /config, and the authorization logic used prefix matching, which inadvertently allowed access to /config/update.

This access control bypass enabled multiple attack chains. Attackers could modify the proxy’s runtime configuration to read arbitrary files, take over administrative accounts, and achieve remote code execution by loading malicious Python modules from cloud storage.

A separate SSRF vulnerability in the Clientside LLM Credentials” feature allowed attackers to redirect API calls and exfiltrate provider keys.

The researchers tested their findings against version 1.77.7. The LiteLLM project resolved all reported issues in version 1.83.9.

What you should do

Organizations running LiteLLM gateway should upgrade to version 1.83.9 or later immediately. The vulnerabilities affect the gateway component specifically, which is commonly deployed as a standalone service in LLM-based applications.

Administrators should review access logs for any unauthorized configuration changes or suspicious API activity during the period when vulnerable versions were deployed. Any API keys or credentials that may have been exposed should be rotated.

What it means

LLM gateways occupy a critical position in AI infrastructure, sitting between applications and model providers while handling credentials, configuration, and request routing. As these systems see wider adoption, they become attractive targets for attackers seeking to compromise AI deployments.

The LiteLLM vulnerabilities demonstrate that rapid evolution in AI infrastructure tools can lead to gaps in foundational security practices. The missing authorization check and overly permissive route matching are classic web application security issues, but their impact is amplified in the context of LLM gateways due to the sensitive data and broad access these systems handle.

The ability to modify configuration at runtime, while useful for operational flexibility, creates significant security risks when not properly protected. In this case, a single authorization bypass enabled multiple high-impact attack chains, from credential theft to remote code execution.

Organizations deploying LLM gateways should ensure that proper access controls are in place, that administrative functions require explicit role checks, and that route matching logic does not inadvertently expose sensitive endpoints. Regular security assessments of AI infrastructure components are essential as the ecosystem continues to evolve.

Sources

Christopher Ort

Christopher Ort

Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU). He is the author of “The Style Penalty”, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.

Editor's picks

  1. CVE-2026-35029: LiteLLM Gateway Authorization Bypass Enables RCE
  2. Cursor marketplace breach traced to unredacted credentials in stored agent traces
  3. AI-Powered Antidetect Browsers Enable Mass Social Media Fraud and Influence Operations
  4. Aokigahara Infostealer Targets AI Tools and Developer Credentials
  5. Yoba malware campaign abuses signed ChatGPT and Claude binaries via DLL sideloading