CVE-2026-25960: Vllm Server Side Request Forgery (CVSS 7.1)

vLLM versions 0.22.0 through 0.23.0 fail to validate stop_token_ids against vocabulary bounds in Rust HTTP and gRPC frontends, allowing out-of-vocabulary token IDs to reach MinTokensLogitsProcessor. Attackers can submit requests with min_tokens greater than zero and out-of-vocabulary stop_token_ids to trigger CUDA tensor indexing failures that leave EngineCore in a fatal state requiring service restart.

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Prior to 0.24.0, the structured_outputs.regex API parameter passes a user-supplied regular expression string directly to the grammar compiler backends with no compilation timeout; in the xgrammar backend the string reaches the regex compiler with no guard, and in the outlines backend the validation step blocks structural issues such as lookarounds and backreferences but performs no complexity analysis, so a pattern with nested quantifiers passes all checks and causes exponential state-space expansion, allowing a single request containing an adversarial regex to hang an inference worker indefinitely and deny service. This issue is fixed in version 0.24.0.

vLLM versions >= 0.10.2 and < 0.13.0 are missing sparse tensor validation in multimodal embeddings processing. Because PyTorch disables sparse tensor invariant checks by default, an attacker can submit crafted embedding requests with malformed (negative or out-of-bounds) tensor indices, when the prompt-embeds feature is enabled, to trigger crashes or resource exhaustion (denial of service), with potential for out-of-bounds/write-what-where memory corruption. This continues CVE-2025-62164, whose prior fix only disabled the feature by default rather than addressing the root cause.

vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.

FastGPT is an AI Agent building platform. Prior to version 4.14.9.5, the FastGPT HTTP tools testing endpoint (/api/core/app/httpTools/runTool) is exposed without any authentication. This endpoint acts as a full HTTP proxy — it accepts a user-supplied baseUrl, toolPath, HTTP method, custom headers, and body, then makes a server-side HTTP request and returns the complete response to the caller. This issue has been patched in version 4.14.9.5.

FastMCP is a Pythonic way to build MCP servers and clients. Prior to version 3.2.0, the OpenAPIProvider in FastMCP exposes internal APIs to MCP clients by parsing OpenAPI specifications. The RequestDirector class is responsible for constructing HTTP requests to the backend service. A vulnerability exists in the _build_url() method. When an OpenAPI operation defines path parameters (e.g., /api/v1/users/{user_id}), the system directly substitutes parameter values into the URL template string without URL-encoding. Subsequently, urllib.parse.urljoin() resolves the final URL. Since urljoin() interprets ../ sequences as directory traversal, an attacker controlling a path parameter can perform path traversal attacks to escape the intended API prefix and access arbitrary backend endpoints. This results in authenticated SSRF, as requests are sent with the authorization headers configured in the MCP provider. This issue has been patched in version 3.2.0.