CVE-2026-24779

vLLM is an inference and serving engine for large language models (LLMs). Prior to version 0.14.1, a Server-Side Request Forgery (SSRF) vulnerability exists in the `MediaConnector` class within the vLLM project's multimodal feature set. The load_from_url and load_from_url_async methods obtain and process media from URLs provided by users, using different Python parsing libraries when restricting the target host. These two parsing libraries have different interpretations of backslashes, which allows the host name restriction to be bypassed. This allows an attacker to coerce the vLLM server into making arbitrary requests to internal network resources. This vulnerability is particularly critical in containerized environments like `llm-d`, where a compromised vLLM pod could be used to scan the internal network, interact with other pods, and potentially cause denial of service or access sensitive data. For example, an attacker could make the vLLM pod send malicious requests to an internal `llm-d` management endpoint, leading to system instability by falsely reporting metrics like the KV cache state. Version 0.14.1 contains a patch for the issue.

An attacker with standard user-level access to a multi-tenant vLLM deployment crafts an image URL that appears to target an allowed external host (e.g., cdn.allowed-host.com) but uses a backslash in the URL path component. The two Python URL parsing libraries used by MediaConnector interpret this differently: the allowlist check sees cdn.allowed-host.com, while the actual HTTP request resolves to an internal address — e.g., http://169.254.169.254/latest/meta-data/ (AWS IMDS) or http://10.96.0.1/api/v1/secrets (Kubernetes API). In an llm-d cluster, the attacker targets the llm-d management endpoint and submits falsified KV cache metrics, causing the scheduler to misroute requests or trigger unnecessary pod scaling, resulting in service degradation across the inference fleet.