A critical security vulnerability has been disclosed in LiteLLM, the widely used open-source proxy framework for managing large language model (LLM) API calls.
Tracked as CVE-2026-49468 (GHSA-4xpc-pv4p-pm3w), the flaw allows unauthenticated attackers to bypass authentication controls and gain unauthorized access to protected management routes via a Host header injection vulnerability.
The vulnerability, classified under CWE-290 (Authentication Bypass by Spoofing), resides in the authentication utility layer of the LiteLLM proxy. Specifically, the flaw exists in litellm/proxy/auth/auth_utils.py within the get_request_route() function.
Critical LiteLLM Vulnerability
This function derives the effective route from request.url.path, which the Starlette web framework reconstructs using the incoming HTTP Host header.
Because the authentication gate evaluates a route path derived from attacker-controlled input, a crafted Host header can cause the auth layer to evaluate an entirely different route than the one FastAPI actually dispatches.
This mismatch creates a critical blind spot: the authentication check passes for a safe-seeming route, while the request is internally routed to a restricted management endpoint.
The vulnerability carries a CVSS v4 base score rated as Critical, with the full vector string CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H.
The attack is network-accessible, requires no privileges, demands no user interaction, and results in a high-impact compromise of confidentiality, integrity, and availability in both the vulnerable system and subsequent systems it interfaces with. All LiteLLM versions prior to 1.84.0 are affected.
However, the advisory notes that most production deployments may not be directly exposed, as the bypass can be blocked by any upstream network layer that validates or normalizes the Host header before forwarding traffic to the proxy.
Deployments protected by the following are not at risk:
- A CDN or Web Application Firewall (WAF) such as Cloudflare
- A reverse proxy configured with explicit
server_nameallowlists - A host-based load balancer with routing rules
- LiteLLM Cloud (managed service customers are confirmed unaffected)
Despite these mitigations, self-hosted and direct-exposure deployments without such upstream controls remain highly vulnerable.
Patch and Remediation
The vulnerability was patched in LiteLLM version 1.84.0. Maintainers strongly recommend that all users upgrade immediately. No configuration changes are required alongside the update the fix is applied automatically upon upgrading to the patched release.
For organizations unable to upgrade immediately, the recommended interim mitigation is to place the LiteLLM proxy behind an upstream component that validates the Host header.
Acceptable options include a CDN/WAF, a reverse proxy with explicit allowlists, or a cloud load balancer with host-based routing enforcement. Restricting network access to the proxy listener at the firewall level is also advisable.
The advisory was published to the GitHub Advisory Database and reviewed by the BerriAI security team.
Given LiteLLM’s growing adoption in enterprise AI infrastructure as a unified gateway for LLM providers, organizations should treat this patch as a high priority and audit their deployment architecture accordingly.
Follow us on Google News , LinkedIn and X to Get More Instant Updates. Set Cyberpress as a Preferred Source in Google.
The post Critical LiteLLM Flaw Enables Authentication Bypass via Host Header Injection appeared first on Cyber Security News.
Discover more from RSS Feeds Cloud
Subscribe to get the latest posts sent to your email.
