PHP2Shell : See if you’re affected with a free external attack surface scan. Click Here > 

Back
New CVE Detected

CVE-2026-94626 – Memory Exhaustion / Denial of Service – vLLM ≤ 0.29.0 (Disaggregated Prefill/Decode

Be the first to know when new zero-days emerge:

Subscribe to Email Alerts
Subscribe to RSS Feed

Summary

CVE-2026-94626 is a memory exhaustion (uncontrolled memory allocation) vulnerability affecting vLLM through version 0.29.0. The flaw arises because vLLM’s OpenAI-compatible completion endpoints fail to validate the tp_size parameter inside kv_transfer_params, allowing a remote, unauthenticated attacker to trigger unbounded memory allocation on decode worker processes in prefill/decode (disaggregated) deployments. The issue is rated HIGH severity and can result in kernel out-of-memory (OOM) kills of inference workers, disrupting service availability.

Technical details

  • Root cause: vLLM does not bound-check or sanity-validate the tp_size value supplied in kv_transfer_params before using it to construct KV-transfer topology and NIXL metadata structures (e.g., in TransferTopology.handshake_target_ranks).
  • Trigger conditions: The target deployment must use vLLM’s disaggregated prefill/decode architecture with a KV-transfer connector (e.g., NIXL) that accepts kv_transfer_params from client requests via the OpenAI-compatible completions API.
  • Attack vector: Network-based, unauthenticated — an attacker sends a crafted completion request specifying an arbitrarily large tp_size value in kv_transfer_params.
  • Impact: Excessive/unbounded memory allocation on the decode worker leads to kernel OOM-killing of the decode EngineCore process, taking the inference worker out of service (denial of service / availability impact only — confidentiality and integrity are not affected).

Affected software

  • vllm-project/vllm (PyPI package vllm) — versions 0 through 0.29.0 (inclusive), specifically deployments running the disaggregated prefill/decode architecture with KV-transfer connectors (e.g., NIXL) exposed via OpenAI-compatible completion endpoints.

Severity

  • CVSS v3.1 Base Score: 7.5 (HIGH) — CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
  • CVSS v4.0 Score: 8.7 (HIGH) — 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
  • CWE-789: Memory Allocation with Excessive Size Value

Mitigation and recommended actions

  • Immediate: Track the upstream fix in vLLM pull request #51137, which introduces defense-in-depth validation of tp_size (and related pp_size/remote_port values) at the API boundary (Pydantic validators on CompletionRequest, ChatCompletionRequest, and ResponsesRequest), inside the KV connector topology logic, and in the request-rejection cleanup path. Upgrade to a vLLM release that incorporates this fix as soon as it is published by the vLLM project.
  • If no patch is yet available:
    • Restrict network access to vLLM OpenAI-compatible completion endpoints to trusted clients only (e.g., via network segmentation, VPN, or API gateway authentication) — do not expose disaggregated prefill/decode endpoints directly to untrusted networks.
    • Where possible, disable or tightly control client-supplied kv_transfer_params (particularly tp_size) on public-facing endpoints.
    • Monitor decode worker memory usage and set resource limits/cgroup constraints on worker processes to reduce the blast radius of an OOM condition.
    • Watch the official vLLM GitHub Security Advisories page for the finalized advisory and patched release version.

References

Are you exposed?

Get a free report of your organization’s exposure to this CVE and threat

  • Mapping of all Assets with this Technology
  • Identification of potentially exposed assets to this CVE
  • Confirmation of verified exploitable assets
Get Exposure Report

How IONIX’s External Exposure Management Platform Detects and Validates
Zero-Days to Shrink MTTR

1

Map your entire attack surface (continously)

IONIX uses multi-factor discovery methods, including DNS analysis, certificate mapping, metadata inspection, and more, to automatically map every internet-facing asset across your environment. This includes cloud instances, third-party platforms, shadow IT, and even forgotten infrastructure that traditional tools miss.

2

Monitor for new CVEs

Dozens of threat intel feeds using agentic technology are continuously analyzed to detect the appearance of proof-of-concept code, exploit kits, and indicators of active targeting. IONIX goes further by applying AI to proactively evaluate whether emerging vulnerabilities are likely to be exploited, even before PoCs go public.

3

Identify Potential External Exposures

Not all CVEs matter. IONIX filters vulnerabilities by asking attacker-centric questions: Can it be reached from the internet? Does it require authentication? Is it being exploited in the wild? This dramatically reduces noise and focuses teams on threats that can actually be weaponized.

4

Create Safe, Scalable Exploit Validations

IONIX transforms real-world PoCs into safe, non-intrusive test payloads that can be run in production environments without disruption. These simulations are precisely targeted to the systems that are vulnerable, ensuring rapid validation without unnecessary load.

5

Execute Exploit Validations

By combining context about software stack, versioning, exposure status, and reachability, IONIX ensures that only the right payloads are executed against the right assets, maximizing efficiency and minimizing risk.

6

Drive Fast and Actionable Remediation

Results are routed through integrations with ticketing, SOAR, and SIEM tools. Issues are written in plain language, bundled into remediation clusters, and prioritized based on asset criticality, exploitability, and blast radius. This shortens mean time to remediation (MTTR) and empowers teams to act with confidence.

Are you exposed?

Get a free report of your organization’s exposure to this CVE and threat

  • Mapping of all Assets with this Technology
  • Identification of potentially exposed assets to this CVE
  • Confirmation of verified exploitable assets
Get Exposure Report
Close

Subscribe to Threat Center RSS

Copy/paste the link below into your preferred RSS reader or follow these instructions to subscribe to Slack alerts.

Open RSS feedSlack setup instructions

Close

Get Real-Time CVE Alerts to Your Email

Be the first to know when new zero-days emerge