{"id":"CVE-2026-116181","kind":"vulnerability","aliases":["CVE-2026-116181"],"related":[],"title":"vLLM: out-of-vocabulary stop_token_ids crash the engine remotely","description":"vLLM from 0.20.1 before 0.29.0 contains an improper input validation vulnerability that allows remote attackers to crash the engine by supplying out-of-vocabulary stop_token_ids. Attackers can send OpenAI chat, completion, generate, or batch requests with invalid IDs to terminate the engine or write outside the V2 runner's logits buffer.","summary":"vLLM versions from 0.20.1 before 0.29.0 fail to validate stop_token_ids in requests, so an out-of-vocabulary ID can crash the inference engine. The same malformed input can also write outside the V2 runner's logits buffer.","severity":"high","cvss":7.1,"epss":null,"exploited":false,"exploited_at":null,"products":[],"ecosystems":[],"tags":["ai-tooling","llm-serving","python","self-hosted","vllm"],"not_affected_if":["default-config"],"how_to_check":"- Check the installed version with `pip show vllm` or `python -c \"import vllm; print(vllm.__version__)\"`\n- Confirm whether the OpenAI-compatible server is reachable from untrusted networks, since the sources say remote attackers can send chat, completion, generate or batch requests","fix":"- Upgrade vLLM to `0.29.0` or later\n- Until upgraded, restrict network access to the vLLM API server","status":"open","published_at":"2026-10-25T14:16:57+00:00","first_seen_at":"2026-10-25T14:41:23+00:00","triaged_at":"2026-10-25T14:42:35+00:00","updated_at":"2026-10-25T14:42:35+00:00","category":"ai-tooling","confidence":0.6,"triage":{"at":"2026-10-25T14:42:35Z","model":"claude-haiku-5-5","output":{"fix":"- Upgrade vLLM to `0.29.0` or later\n- Until upgraded, restrict network access to the vLLM API server","reach":"internet","title":"vLLM: out-of-vocabulary stop_token_ids crash the engine remotely","impact":"high","summary":"vLLM versions from 0.20.1 before 0.29.0 fail to validate stop_token_ids in requests, so an out-of-vocabulary ID can crash the inference engine. The same malformed input can also write outside the V2 runner's logits buffer.","category":"ai-tooling","nicknames":[],"who_cares":["vllm","python","self-hosted","llm-serving"],"confidence":0.6,"needs_auth":"unknown","worst_case":"Remote requests crash the inference engine and may corrupt memory outside the logits buffer","analyst_note":"Sources do not say whether authentication is required or if the V2 runner is default; check the NVD entry and vLLM 0.29.0 release notes for the fix details.","how_to_check":"- Check the installed version with `pip show vllm` or `python -c \"import vllm; print(vllm.__version__)\"`\n- Confirm whether the OpenAI-compatible server is reachable from untrusted networks, since the sources say remote attackers can send chat, completion, generate or batch requests","needs_analyst":true,"not_affected_if":["default-config"],"who_is_affected":"Teams running vLLM between versions 0.20.1 and 0.29.0 that expose its OpenAI-compatible endpoints to clients they do not fully trust. Internal deployments with no untrusted callers are lower exposure. unknown whether the V2 runner is the default path in every configuration.","user_interaction":"no","needs_special_setup":"unknown","severity_if_unknown":"high"},"cost_usd":0.000642,"input_tokens":3346,"output_tokens":615,"severity_before":"high"},"analyst_status":null,"analyst_note":null,"search":"'-116181':15A,18A '-2026':14A,17A '0.20.1':33B,72C '0.29.0':35B,74C 'ai':20B 'ai-tool':19B 'allow':82C 'also':61B 'attack':84C,98C 'batch':106C 'buffer':69B,123C 'chat':102C 'complet':103C 'contain':75C 'crash':9A,52B,86C 'cve':13A,16A 'engin':11A,55B,88C,114C 'fail':36B 'generat':104C 'host':28B 'id':8A,41B,50B,97C,110C 'improp':77C 'infer':54B 'input':59B,78C 'invalid':109C 'llm':23B 'llm-serv':22B 'logit':68B,122C 'malform':58B 'openai':101C 'out-of-vocabulari':2A,46B,91C 'outsid':63B,117C 'python':25B 'remot':12A,83C 'request':43B,107C 'runner':66B,120C 'self':27B 'self-host':26B 'send':100C 'serv':24B 'stop':6A,39B,95C 'suppli':90C 'termin':112C 'token':7A,40B,96C 'tool':21B 'v2':65B,119C 'valid':38B,79C 'version':31B 'vllm':1A,29B,30B,70C 'vocabulari':5A,49B,94C 'vulner':80C 'write':62B,116C","nicknames":[],"mitigations":[],"who_affected":"Teams running vLLM between versions 0.20.1 and 0.29.0 that expose its OpenAI-compatible endpoints to clients they do not fully trust. Internal deployments with no untrusted callers are lower exposure. unknown whether the V2 runner is the default path in every configuration.","tier":"B","verified":{},"analyst_revisit_at":null,"first_public_at":"2026-10-25T14:16:57+00:00","kev":null,"cvss_vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X","cvss_by":"disclosure@vulncheck.com","poc_count":0,"sources":[{"source":"nvd","external_id":"CVE-2026-116181","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-116181","severity":null,"published_at":"2026-10-25T14:16:57+00:00","modified_at":"2026-10-25T14:16:57+00:00","first_seen_at":"2026-10-25T14:41:23+00:00","fetched_at":"2026-10-25T14:41:23+00:00","note":null}],"documents":[],"notice":"Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai"}