NetSight

Security Dashboard

A live picture of new vulnerabilities, flaws exploited in the wild and malicious packages.

Reset
159 matching advisories · page 3 of 6Index updated 3 minutes ago · Subscribe · RSS
SeverityIDTitleAffectedAge
high7.5 CVE-2026-94627 vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in prefi… vllm · vllm 2026-09-21 high7.5 CVE-2026-94626 vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints, allowing attackers to allocate unb… vllm · vllm 2026-09-21 medium5.3 CVE-2026-94625 vLLM through 0.29.0 contains a resource exhaustion vulnerability in MooncakeConnector where rejected prefill requests create ownerless transfer placeholders th… vllm · vllm 2026-09-21 high7.5 CVE-2026-94624 vLLM through 0.29.0 contains a denial of service vulnerability in P2P KV offloading when OffloadingConnector is configured with TieringOffloadingSpec and a pee… pip · vllm 2026-09-21 high7.5 CVE-2026-94623 vLLM through 0.29.0 contains a denial of service vulnerability in the NIXL connector's prefix caching implementation that fails to properly validate block coun… vllm · vllm 2026-09-21 high7.5 CVE-2026-94622 vLLM versions through 0.29.0 contain a denial of service vulnerability in the NIXL connector's metadata handling for prefill/decode disaggregated deployments. vllm · vllm 2026-09-21 low3.1 CVE-2026-93989 vLLM through 0.29.0 fails to properly validate bad_words token indices against the model's generation output width in SamplingParams.update_from_tokenizer(). vllm · vllm 2026-09-19 low3.7 CVE-2026-93841 vLLM through 0.29.0 contains a memory corruption vulnerability in the Triton _bincount_kernel where prompt token IDs index the penalty prompt-presence bitset w… vllm · vllm 2026-09-18 low3.7 CVE-2026-93840 vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead of model output logits width in SamplingParams._validate_allowed_token_ids(). vllm · vllm 2026-09-18 high7.5 CVE-2026-93592 vLLM versions before 0.28.0 fail to validate the lower bound of token IDs in the /v1/embeddings and /pooling endpoints, allowing unauthenticated attackers to c… vllm · vllm 2026-09-18 high7.5 CVE-2026-93436 vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. vllm · vllm 2026-09-17 medium6.5 CVE-2026-69147 vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservation pip · vllm 2026-09-16 medium6.5 CVE-2026-57173 vLLM: Unauthenticated audio decompression-bomb DoS in /v1/chat/completions pip · vllm 2026-09-16 medium4.3 CVE-2026-92365 A vulnerability was found in vllm-project vllm up to 0.29.0. vllm-project vllm 2026-09-16 medium5.3 CVE-2026-92220 A vulnerability was found in vllm-project vLLM 0.26.0/0.27.0. 2026-09-16 medium4.3 CVE-2026-90878 A vulnerability was determined in vllm-project vLLM up to 0.27.1. vllm-project vLLM 2026-09-15 low3.3 CVE-2026-90713 A security flaw has been discovered in vllm-project vLLM up to 0.29.0. vllm-project vLLM 2026-09-14 medium6.5 CVE-2026-90555 vLLM versions before 0.28.0 fail to validate audio sample rate headers in the transcription endpoint, allowing authenticated clients to bypass duration checks. vllm · vllm 2026-09-12 medium6.2 CVE-2026-90554 vLLM versions >=0.10.2 and <0.28.0 do not apply any audio decode-size or duration limit when extracting audio from video input for NanoNemotronVL models. vllm · vllm 2026-09-12 high7.8 CVE-2026-90553 vLLM before 0.28.0 contains a remote code execution vulnerability in the LlavaOnevision2 processor loader that ignores the trust_remote_code parameter when loa… vllm · vllm 2026-09-12 high7.5 CVE-2026-37237 vLLM up to and including 0.17.0 allows remote attackers to cause a Denial of Service via memory exhaustion. vllm · vllm 2026-08-28 medium5.3 CVE-2026-78684 vLLM before 0.27.0 fails to properly classify DeepStream as a GPU backend and omits pixel-limit enforcement in its decode path. vllm · vllm 2026-08-25 medium6.5 CVE-2026-73560 vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protections pip · vllm 2026-08-17 medium4.3 CVE-2026-71486 vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds pip · vllm 2026-08-17 medium6.5 CVE-2026-73559 vLLM: Completion prompt lists fan out into unbounded engine requests pip · vllm 2026-08-13 medium5.3 CVE-2026-73558 vLLM is an inference and serving engine for large language models. pip · vllm 2026-08-13 medium5.3 CVE-2026-73557 vLLM: Incomplete CVE-2025-62164 remediation can be bypassed by concurrent prompt parts pip · vllm 2026-08-13 medium5.3 CVE-2026-73556 vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2m pip · vllm 2026-08-13 medium5.3 CVE-2026-73555 vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error Messages pip · vllm 2026-08-13 medium5.5 CVE-2026-27765 Improper input validation for some vLLM Hardware Plugin for Intel(R) Gaudi(R) software before version 0.16.0 within Ring 3: User Applications may allow a denia… intel · vllm_hardware 2026-08-11

Get notifications

Subscribe to new vulnerabilities and early warnings from the last seven days. Pick what matters to your stack and copy the address into any feed reader. More filters appear when you pick a type.

Format:
Open feed

Early warnings are advisories that are published before or without a CVE number, for example malicious packages, advisories with a CVE still pending and reports that are not reviewed yet. Confidence says how well an entry is checked: high, medium or low.

About this dashboard

The dashboard keeps its own index of security advisories and refreshes it every 30 minutes. It merges the national vulnerability database, package advisories, malware reports and the catalog of vulnerabilities that are exploited in the wild into one record per issue. Reports about the same flaw are combined, so nothing is listed twice.

Open any entry for the full record: description, severity vector, affected packages and versions, fixed versions, weakness class, exploitation probability and references.

Sources: the GitHub Advisory Database (reviewed advisories, malware reports and unreviewed reports), the National Vulnerability Database, the CVE Program (cve.org) for brand-new records, the OpenSSF malicious packages list, the CISA catalog of Known Exploited Vulnerabilities and Microsoft's blocklist of malicious VS Code extensions. Every entry names its sources on its detail page.

This product uses the NVD API but is not endorsed or certified by the NVD.

Reading the figures

  • Threat level: a rating computed from three numbers shown next to it: critical advisories in the last 24 hours, vulnerabilities newly confirmed as exploited this week and malware reports this week. It is a quick orientation, not an official warning level.
  • Severity: scores from 9.0 are critical, 7.0 to 8.9 high, 4.0 to 6.9 medium and 0.1 to 3.9 low. Severity describes how bad a flaw is in general. It does not know whether the component is reachable in your environment.
  • Known exploited: exploitation has been observed in real attacks. These are the entries to patch first. Entries with a fix deadline show how many days are left.
  • Early warnings: advisories that appear before or without a CVE number, for example malicious packages, reviewed advisories with a CVE still pending and reports that are not reviewed yet.
  • Priority queue: critical and high advisories of the last 7 days and everything newly exploited, ordered by exploitation status, severity and score.

Questions

Where does the data come from?
From public vulnerability databases, package advisories and the catalog of exploited vulnerabilities. NetSight collects them into its own index every 30 minutes and merges duplicates.
How do I check whether I am affected?
Search for your product or package name, open the record and compare your installed version with the affected and fixed versions. A product name alone does not mean your version is affected.
What does known exploited mean?
Exploitation of this vulnerability has been observed in real attacks. Such entries carry a date they were confirmed and, where one exists, a deadline for applying the fix.
What is an early warning?
An advisory that is published before or without a CVE number, such as a malicious package or a reviewed advisory whose CVE is still pending. They give you a head start, but the details can change.
Is a high severity always more urgent than a low one?
No. Severity measures technical impact in general. Exposure of the affected system, observed exploitation and the exploitation probability matter as much.
Can I follow new advisories automatically?
Yes. Use the notification builder further down to create an RSS, Atom or JSON feed filtered by severity, technology, ecosystem or exploitation status.