vLLM security advisories

114 advisories · 34 critical or high in 12 months · latest Oct 28

60 of 114 advisories

vLLM: Permissive default CORS origins allow cross-origin requests to the vLLM HTTP serverCVE-2026-115917fixed in 0.32.0 Medium 6.5Oct 28

vLLM: resource exhaustionCVE-2026-115913fixed in 0.32.0 Medium 6.5Oct 28

vLLM DeepSeek-OCR image processing vulnerable to denial of serviceCVE-2026-115914fixed in 0.32.0 High 8.1Oct 28

vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCECVE-2026-115915fixed in 0.32.0 High 7.5Oct 28

vLLM: unsafe deserializationCVE-2026-115912fixed in 0.32.0 Medium 6.5Oct 28

vLLM: improper array index validationCVE-2026-115916fixed in 0.32.0 Medium 6.5Oct 28

vLLM: Responses API bypasses trust_request_chat_template for chat_template_kwargsCVE-2026-115918fixed in 0.32.0 Medium 6.5Oct 28

vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread poolCVE-2026-115911fixed in 0.32.0 High 8.1Oct 27

vLLM: Request-controlled mm_processor_kwargs.code_revision allows remote code executionCVE-2026-114053fixed in 0.31.0 Medium 6.5Oct 27

vllm p2p never connected zmq_exhaustionCVE-2026-94624fixed in 0.32.0 High 8.2Oct 25

vLLM: server-side request forgeryCVE-2026-116176fixed in 0.24.0 High 7.5Oct 25

vLLM: resource exhaustionCVE-2026-116177fixed in 0.25.0 High 7.5Oct 25

vLLM: denial of serviceCVE-2026-116178fixed in 0.30.0 High 7.5Oct 25

vLLM: denial of serviceCVE-2026-116173fixed in 0.24.0 High 7.5Oct 23

vLLM : DoS via unvalidated mm_processor_kwargs in multimodal image processors (Isaac, InternVL, H2OVL, etc.)CVE-2026-114055fixed in 0.31.0 Medium 6.5Oct 23

vLLM: insufficient authenticity checkCVE-2026-114050fixed in 0.31.0 Medium 6.5Oct 21

vLLM: authenticated attacker could collide shared cache keysCVE-2026-114054fixed in 0.31.0 High 7.5Oct 21

vLLM after 0.24.0 and before 0.31.0 contains a memory exhaustion vulnerability...CVE-2026-114052fixed in 0.31.0 Medium 4.3Oct 6

vLLM: denial of serviceCVE-2026-105922fixed in 0.32.0 Medium 6.5Oct 6

vLLM: reachable assertionCVE-2026-105753fixed in 0.28.0 Low 3.1Oct 6

vLLM: Harmony tool continuations drop `cache_salt` , restoring a cross-tenant prefix-cache membership oracleCVE-2026-105752fixed in 0.30.0 Medium 5.3Oct 5

vLLM: resource exhaustionCVE-2026-105758fixed in 0.30.0 Medium 5.3Oct 5

vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustionCVE-2026-105760fixed in 0.30.0 Medium 6.5Oct 5

vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied featuresCVE-2026-105754fixed in 0.30.0 Medium 6.5Oct 5

vLLM: improper input validationCVE-2026-105757fixed in 0.30.0 Medium 4.2Oct 5

vLLM: insecure direct object referenceCVE-2026-105755fixed in 0.30.0 Medium 6.5Oct 5

vLLM: improper input validationCVE-2026-105756fixed in 0.30.0 High 7.5Oct 5

vLLM: Rust frontend's track_http_metrics middleware records the raw HTTP method...CVE-2026-105759fixed in 0.30.0 High 7.5Sep 30

vLLM: denial of serviceCVE-2026-103241fixed in 0.27.0 Medium 6.5Sep 26

vLLM through 0.29.0 fetches and fully materializes remote or inline media...CVE-2026-100650fixed in 0.30.0 High 7.5Sep 26

vLLM versions 0.22.0 through 0.23.0 fail to validate stop_token_ids against...CVE-2026-100652fixed in 0.24.0 Medium 6.5Sep 26

vLLM: operator-supplied model revision pin is not propagated to several Hugging...CVE-2026-100653fixed in 0.28.0 High 7.5Sep 21

vLLM: denial of serviceCVE-2026-94623fixed in 0.30.0 High 7.5Sep 21

vLLM: denial of serviceCVE-2026-94624fixed in 0.30.0 Medium 5.3Sep 21

vLLM: resource exhaustionCVE-2026-94625fixed in 0.30.0 High 7.5Sep 21

vLLM: attacker could allocate unbounded memoryCVE-2026-94626fixed in 0.30.0 High 7.5Sep 21

vLLM: denial of serviceCVE-2026-94622fixed in 0.30.0 Medium 4.3Sep 19

vLLM through 0.29.0 fails to properly validate bad_words token indices against...CVE-2026-93989fixed in 0.30.0 Medium 5.3Sep 18

vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead...CVE-2026-93840fixed in 0.29.0 Medium 5.3Sep 18

vLLM: memory corruptionCVE-2026-93841fixed in 0.30.0 High 7.5Sep 18

vLLM: unauthenticated attacker could crash the engineCVE-2026-93592fixed in 0.28.0 High 7.5Sep 17

vLLM through 0.29.0 fails to properly clean up decode-side metadata for...CVE-2026-93436fixed in 0.30.0 Medium 6.5Sep 17

vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservationCVE-2026-69147fixed in 0.28.0 Medium 6.5Sep 16

vLLM: Unauthenticated audio decompression-bomb DoS in /v1/chat/completionsCVE-2026-57173fixed in 0.24.0 High 7.8Sep 12

vLLM: remote code executionCVE-2026-90553fixed in 0.28.0 Medium 6.5Sep 8

vLLM: server-side request forgeryCVE-2026-73560fixed in 0.26.0 Medium 5.3Sep 8

vLLM: Cross-User Data Leak VulnerabilityCVE-2026-73558fixed in 0.27.0 MediumSep 4

vLLM: Incomplete CVE-2025-62164 remediation can be bypassed by concurrent prompt partsCVE-2026-73557fixed in 0.26.0 Medium 5.3Sep 4

vLLM: denial of serviceCVE-2026-73556fixed in 0.26.0 Medium 5.3Sep 4

vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error MessagesCVE-2026-73555fixed in 0.26.0 Medium 4.3Sep 4

vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output boundsCVE-2026-71486fixed in 0.26.0 Medium 5.3Aug 25

vLLM: denial of serviceCVE-2026-78684fixed in 0.27.0 Medium 6.5Aug 13

vLLM: Completion prompt lists fan out into unbounded engine requestsCVE-2026-73559fixed in 0.26.0 HighJul 20

vLLM denial of service via prompt embeds on M-RoPE modelsCVE-2026-55514fixed in 0.24.0 Medium 6.5Jul 17

vLLM: Speech-to-text upload size limit is enforced after full UploadFile readCVE-2026-55646fixed in 0.24.0 High 7.5Jul 17

vLLM: ReDoS via structured_outputs.regex compiled without timeout in xgrammar and outlines backendsCVE-2026-55574fixed in 0.24.0 High 7.5Jul 17

vLLM has Remote DoS via Invalid Recovered Token ReinjectionCVE-2026-54234fixed in 0.24.0 Medium 5.9Jul 17

vLLM: improper input validationCVE-2026-34760fixed in 0.18.0 High 8.8Jun 22

vLLM: code executionCVE-2026-54232fixed in 0.22.1 Medium 6.5Jun 17

vLLM: OOM Denial of Service via Audio Decompression BombCVE-2026-54233fixed in 0.24.0

About vLLM

The high-throughput LLM inference and serving engine.

Packages watched: vllm (PyPI).

vllm-project/vllm On Repos