{"id":2475,"cve":"CVE-2026-105757","cve_url":"https:\/\/www.cvedetails.com\/cve\/CVE-2026-105757\/","cwe":"CWE-20","cwe_number":20,"cwe_url":"https:\/\/cwe.mitre.org\/data\/definitions\/20","cvss_score":6.5,"published":"2026-10-05","owasp":"A08 Software and Data Integrity Failures","odc":"Checking","cdc":"Missing Check","category":"Input Validation and Sanitization","subcategory":"Insecure Parsing or Deserialization","scope":"Remote","impact":"Denial of Service (DoS)","summary":"vLLM <0.30.0 allows structured-output bugs to crash the shared engine.","permalink":"https:\/\/netpack.pt\/vaitp\/vulnerability\/CVE-2026-105757\/","description":"vLLM is an inference and serving engine for large language models. Prior to 0.30.0, structured-output request failures can escape request-scoped validation and reach the EngineCore fatal-error path. A per-request backend mismatch can re-raise a grammar compilation exception, padding produced by the ngram_gpu speculative-decoding mode can pass a negative token to guidance validation, and the Rust frontend can admit empty structured-output values that the Python frontend rejects, allowing ordinary constrained-generation requests to terminate the shared engine. This issue is fixed in version 0.30.0.","solution":"Upgrade vLLM to version\u202f0.30.0 or later.","affected_component":"vLLM","fixed_by_upgrading":true,"vulnerable_code":"import json\n\nclass EngineCore:\n    def __init__(self):\n        self.state = {}\n\n    def _validate_structured_output(self, output):\n        # VULNERABLE: does not reject empty structured-output values\n        # Missing validation allows empty dict to propagate\n        return True\n\n    def process_request(self, request_json: str):\n        request = json.loads(request_json)\n        output = request.get(\"structured_output\", {})\n        if self._validate_structured_output(output):\n            # Simulate downstream processing that expects non\u2011empty output\n            result = self._run_generation(output)\n            return result\n\n    def _run_generation(self, output):\n        # Placeholder for generation logic that crashes on empty output\n        if not output:\n            raise RuntimeError(\"Generated empty structured output\")\n        return {\"generated\": \"text\", \"metadata\": output}\n","patched_code":"import json\n\nclass EngineCore:\n    def __init__(self):\n        self.state = {}\n\n    def _validate_structured_output(self, output):\n        # FIX: reject empty structured-output values before processing\n        if not isinstance(output, dict) or not output:\n            raise ValueError(\"structured_output must be a non\u2011empty dict\")\n        return True\n\n    def process_request(self, request_json: str):\n        request = json.loads(request_json)\n        output = request.get(\"structured_output\", {})\n        if self._validate_structured_output(output):\n            result = self._run_generation(output)\n            return result\n\n    def _run_generation(self, output):\n        # Placeholder for generation logic\n        return {\"generated\": \"text\", \"metadata\": output}\n","payload_needed":true,"payload":null,"payload_status":"model_refused","published_fix":[{"repository":"vllm-project\/vllm","commit_url":"https:\/\/github.com\/vllm-project\/vllm\/commit\/c55e15a44ec4127832d4a86928a356fdd9e68dbd","commit_subject":"[Structured Output] Keep invalid structured-output requests from stopping the engine (#51450)","units":[{"file":"vllm\/v1\/core\/sched\/scheduler.py","function":null},{"file":"vllm\/v1\/core\/sched\/scheduler.py","function":"Scheduler.update_draft_token_ids"},{"file":"vllm\/v1\/core\/sched\/scheduler.py","function":"Scheduler.update_draft_token_ids_in_output"},{"file":"vllm\/v1\/structured_output\/backend_guidance.py","function":null},{"file":"vllm\/v1\/structured_output\/backend_guidance.py","function":"GuidanceGrammar.validate_tokens"},{"file":"vllm\/v1\/structured_output\/utils.py","function":null}]}],"references":null}