gcdRadar
archive/Patch now/AIR-2026-1090
closedclosed 22 May 2026Patch now2 Disclosed vulnerabilityAI

CVE-2026-5817: Docker Model Runner vllm-metal arbitrary code execution vulnerability

vllm-metal, Docker Model Runner

The vllm-metal inference backend in Docker Model Runner unconditionally enables trust_remote_code=True without sandboxing. This allows malicious models to execute arbitrary Python code on the host.

disclosed CVENo confirmed exploitation, so this record is tracked as a disclosed vulnerability and not counted as an incident. It moves into the counts if CISA or the vendor confirms exploitation. Why

CVSS
8.8
EPSS
<0.01 · top 87% · 2026-11-13
EPSS scores by FIRST
fix status
Fix status unknown
first report
22 May 2026
last update
22 May 2026
CVE ids
CVE-2026-5817other records

Timeline · 1

  1. 22 May 2026CVE-2026-5817: The vllm-metal inference backend in Docker Model Runner on macOS unconditionally sets trust_remote_code=True when loading model…

Sources · 1

Advisories and databases

  1. 2026-05-22
    CVE-2026-5817: The vllm-metal inference backend in Docker Model Runner on macOS unconditionally sets trust_remote_code=True when loading model…
    NVD · nvd.nist.gov/vuln/detail/CVE-2026-5817

Titles and links belong to their publishers. The summary above is written by gcdRadar in its own words; data is CC BY-SA 4.0.

Classification

lane
Patch now
AI role
Attack on AI
categories
Infrastructure vulnerabilitySupply chain
severity
2 · Disclosed vulnerability
fix status
Fix status unknown
targets
Framework / tooling
owasp llm top 10
LLM03:2025 Supply Chain
mitre atlas
AML.T0010 AI Supply Chain Compromise
tags
remote-code-executiontrust-remote-codesupply-chain
dates
happened 22 May 2026 · first reported 22 May 2026
origin
NVD
Vulnerability record from NVD / the GitHub Advisory Database. This product uses the NVD API but is not endorsed or certified by the NVD.
record
confidence 0.95 · created 2026-10-18 10:04Z · updated 2026-10-18 10:04Z

Lane, AI role, categories and severity come from the classifier; the OWASP and ATLAS rows are derived from the categories. How these are assigned

Copy a brief

This record as a short Markdown brief, for a ticket, a chat or a coding agent.

the brief as text
# CVE-2026-5817: Docker Model Runner vllm-metal arbitrary code execution vulnerability

https://gcdradar.vercel.app/incidents/cve-2026-5817-docker-model-runner-vllm-metal-arbitrary-code · AIR-2026-1090 · closed 2026-05-22

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2026-05-22
- AI role: Attack on AI
- Products: vllm-metal, Docker Model Runner
- CVE ids: CVE-2026-5817
- CVSS: 8.8

## Summary

The vllm-metal inference backend in Docker Model Runner unconditionally enables trust_remote_code=True without sandboxing. This allows malicious models to execute arbitrary Python code on the host.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-05-22: CVE-2026-5817: The vllm-metal inference backend in Docker Model Runner on macOS unconditionally sets trust_remote_code=True when loading model… (https://nvd.nist.gov/vuln/detail/CVE-2026-5817)

## Sources

- 2026-05-22 NVD: CVE-2026-5817: The vllm-metal inference backend in Docker Model Runner on macOS unconditionally sets trust_remote_code=True when loading model… https://nvd.nist.gov/vuln/detail/CVE-2026-5817

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.

Related

  1. closed7.5
    vLLM CVE-2026-41523: vLLM Security Check Bypass via assert Statement

    Patch now · Disclosed · since 16 Jun 2026 · updated 22 Jun 2026 · CVE-2026-41523

    AI
  2. closed6.5
    vLLM CVE-2026-47155: vLLM artifact pin decay allows loading unpinned code and weights

    Patch now · Disclosed · since 10 Jun 2026 · updated 22 Jun 2026 · CVE-2026-47155

    AI
  3. closed8.8
    vLLM vLLM Hardcoded Trust Remote Code Parameter Leads to RCE

    Patch now · Disclosed · since 28 May 2026 · CVE-2026-4944

    AI
  4. closed9.8
    Guardrails AI CVE-2026-31233: Guardrails AI code injection in Hub installer

    Supply chain · Disclosed · since 12 May 2026 · CVE-2026-31233

    AI
  5. closed8.2
    LangChain family CVE-2026-44843: LangChain vulnerable to unsafe deserialization

    Patch now · Disclosed · since 8 May 2026 · updated 26 May 2026 · CVE-2026-44843

    AI
  6. closed8.8
    Hugging Face libraries Hugging Face Diffusers custom pipeline trust_remote_code bypass vulnerability

    Patch now · Disclosed · since 7 May 2026 · CVE-2026-44827

    AI