{"id":"CVE-2026-126397","kind":"vulnerability","aliases":["CVE-2026-126397"],"related":[],"title":"vLLM: unbounded image patch setting lets remote clients exhaust CPU","description":"vLLM from 0.10.1 before 0.31.0 forwards a request-controlled mm_processor_kwargs.max_patches value to the Hugging Face processor for Cohere2 Vision models without a server-side upper bound, causing quadratic CPU work during multimodal preprocessing. Remote clients can send small image chat-completion requests with large, varying max_patches values to bypass the LRU cache, exhaust preprocessing workers, and deny service to other clients.","summary":"vLLM forwards a request-supplied max_patches value to the Hugging Face processor for Cohere2 Vision models with no server-side upper bound. A remote client can send small image requests with large varying values to force quadratic CPU work and deny service to other users.","severity":"high","cvss":8.7,"epss":null,"exploited":false,"exploited_at":null,"products":[{"name":"vllm","fixed":[],"affected":"","ecosystem":"pypi"}],"ecosystems":["pypi"],"tags":["ai-tooling","llm-serving","multimodal","self-hosted","vllm"],"not_affected_if":["managed-hosting","no-untrusted-clients","text-only-models"],"how_to_check":"- Check the installed vLLM version with `pip show vllm` and compare it to the range below.\n- Check whether the server serves Cohere2 Vision models that accept image input.","fix":"- `vllm` `0.31.0` or later (fixed version per the NVD range: before 0.31.0 is affected).","status":"open","published_at":"2026-11-13T10:17:43+00:00","first_seen_at":"2026-11-13T10:47:19+00:00","triaged_at":"2026-11-13T10:50:22+00:00","updated_at":"2026-11-13T10:50:22+00:00","category":"ai-tooling","confidence":0.7,"triage":{"at":"2026-11-13T10:50:22Z","model":"claude-haiku-5-5","output":{"fix":"- `vllm` `0.31.0` or later (fixed version per the NVD range: before 0.31.0 is affected).","reach":"network","title":"vLLM: unbounded image patch setting lets remote clients exhaust CPU","impact":"medium","summary":"vLLM forwards a request-supplied max_patches value to the Hugging Face processor for Cohere2 Vision models with no server-side upper bound. A remote client can send small image requests with large varying values to force quadratic CPU work and deny service to other users.","audience":"businesses","category":"ai-tooling","nicknames":[],"who_cares":["vllm","self-hosted","llm-serving","multimodal"],"confidence":0.7,"fix_status":"available","needs_auth":"unknown","worst_case":"Denial of service: preprocessing workers are exhausted and other clients are blocked.","analyst_note":"","how_to_check":"- Check the installed vLLM version with `pip show vllm` and compare it to the range below.\n- Check whether the server serves Cohere2 Vision models that accept image input.","plain_action":"","product_name":"vllm","needs_analyst":false,"not_affected_if":["text-only-models","managed-hosting","no-untrusted-clients"],"who_is_affected":"Teams that run a vLLM server with Cohere2 Vision multimodal models and accept image chat-completion requests from clients they do not fully trust. Text-only deployments and servers that do not serve these models are not exposed by this description.","user_interaction":"no","product_ecosystem":"pypi","needs_special_setup":"yes","severity_if_unknown":"high"},"cost_usd":0.000746,"input_tokens":4676,"output_tokens":556,"severity_before":"high"},"analyst_status":null,"analyst_note":null,"search":"'-126397':13A,16A '-2026':12A,15A '0.10.1':79C '0.31.0':81C 'ai':19B 'ai-tool':18B 'bound':53B,105C 'bypass':130C 'cach':133C 'caus':106C 'chat':120C 'chat-complet':119C 'client':8A,56B,114C,142C 'cohere2':44B,96C 'complet':121C 'control':86C 'cpu':10A,69B,108C 'cve':11A,14A 'deni':72B,138C 'exhaust':9A,134C 'face':41B,93C 'forc':67B 'forward':30B,82C 'host':27B 'hug':40B,92C 'imag':3A,60B,118C 'larg':63B,124C 'let':6A 'llm':22B 'llm-serv':21B 'lru':132C 'max':35B,126C 'mm_processor_kwargs.max':87C 'model':46B,98C 'multimod':24B,111C 'patch':4A,36B,88C,127C 'preprocess':112C,135C 'processor':42B,94C 'quadrat':68B,107C 'remot':7A,55B,113C 'request':33B,61B,85C,122C 'request-control':84C 'request-suppli':32B 'self':26B 'self-host':25B 'send':58B,116C 'serv':23B 'server':50B,102C 'server-sid':49B,101C 'servic':73B,139C 'set':5A 'side':51B,103C 'small':59B,117C 'suppli':34B 'tool':20B 'unbound':2A 'upper':52B,104C 'user':76B 'valu':37B,65B,89C,128C 'vari':64B,125C 'vision':45B,97C 'vllm':1A,17A,28B,29B,77C 'without':99C 'work':70B,109C 'worker':136C","nicknames":[],"mitigations":[],"who_affected":"Teams that run a vLLM server with Cohere2 Vision multimodal models and accept image chat-completion requests from clients they do not fully trust. Text-only deployments and servers that do not serve these models are not exposed by this description.","tier":"B","verified":{},"analyst_revisit_at":null,"fix_status":"available","audience":"businesses","plain_action":null,"pkeys":["vllm"],"ptokens":["vllm"],"labels":["aitooling","high","llmserving","multimodal","pypi","selfhosted","vllm"],"campaign_type":null,"search_a":"'-126397':13A,16A '-2026':12A,15A 'client':8A 'cpu':10A 'cve':11A,14A 'exhaust':9A 'imag':3A 'let':6A 'patch':4A 'remot':7A 'set':5A 'unbound':2A 'vllm':1A,17A","search_b":"'ai':19B 'ai-tool':18B 'bound':53B 'client':56B 'cohere2':44B 'cpu':69B 'deni':72B 'face':41B 'forc':67B 'forward':30B 'host':27B 'hug':40B 'imag':60B 'larg':63B 'llm':22B 'llm-serv':21B 'max':35B 'model':46B 'multimod':24B 'patch':36B 'processor':42B 'quadrat':68B 'remot':55B 'request':33B,61B 'request-suppli':32B 'self':26B 'self-host':25B 'send':58B 'serv':23B 'server':50B 'server-sid':49B 'servic':73B 'side':51B 'small':59B 'suppli':34B 'tool':20B 'upper':52B 'user':76B 'valu':37B,65B 'vari':64B 'vision':45B 'vllm':28B,29B 'work':70B","search_c":"'0.10.1':79C '0.31.0':81C 'bound':105C 'bypass':130C 'cach':133C 'caus':106C 'chat':120C 'chat-complet':119C 'client':114C,142C 'cohere2':96C 'complet':121C 'control':86C 'cpu':108C 'deni':138C 'exhaust':134C 'face':93C 'forward':82C 'hug':92C 'imag':118C 'larg':124C 'lru':132C 'max':126C 'mm_processor_kwargs.max':87C 'model':98C 'multimod':111C 'patch':88C,127C 'preprocess':112C,135C 'processor':94C 'quadrat':107C 'remot':113C 'request':85C,122C 'request-control':84C 'send':116C 'server':102C 'server-sid':101C 'servic':139C 'side':103C 'small':117C 'upper':104C 'valu':89C,128C 'vari':125C 'vision':97C 'vllm':77C 'without':99C 'work':109C 'worker':136C","cause_entry":null,"cause_path":null,"disclosures_checked_at":null,"first_public_at":"2026-10-27T22:20:43+00:00","kev":null,"cvss_vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:N/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-2026-126397","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-126397","severity":null,"published_at":"2026-11-13T10:17:43+00:00","modified_at":"2026-11-13T10:17:44+00:00","first_seen_at":"2026-11-13T10:47:19+00:00","fetched_at":"2026-11-13T10:47:19+00:00","note":null}],"documents":[{"id":145036,"url":"https://github.com/vllm-project/vllm/commit/f6326f53bda46898a331c2d24500332c285d9a2b","title":"Patch: github.com","kind":"other","source":"nvd","publisher":"vendor","text":null,"published_at":null,"relation":"fix","confidence":null,"note":"Patch: github.com"},{"id":110585,"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-9g3c-x747-c6wr","title":"Maintainer's advisory: github.com","kind":"vendor","source":"ghsa-repo","publisher":"vendor","text":null,"published_at":"2026-10-27T22:20:43+00:00","relation":"advisory","confidence":null,"note":"Vendor advisory: github.com"},{"id":145038,"url":"https://www.vulncheck.com/advisories/vllm-0.10.1-before-0.31.0-cohere2-vision-dos-via-max-patches","title":"Third-party advisory: www.vulncheck.com","kind":"advisory","source":"nvd","publisher":"research","text":null,"published_at":null,"relation":"advisory","confidence":null,"note":"Third-party advisory: www.vulncheck.com"}],"disclosures":[],"exploit_sources":[],"notice":"Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai"}