High

Remote Code Execution in vLLM via trust_remote_code Bypass

IdentifiersCVE-2026-27893CWE-693· Protection Mechanism Failure

CVE-2026-27893 affects vLLM, an inference and serving engine for large language models. In versions starting at 0.10.1 and prior to 0.18.0, two model implementation files hardcode trust_remote_code=True when loading sub-components. This overrides and bypasses the user's explicit --trust-remote-code=False setting, defeating the intended security control that disables execution of remote model code. As a result, loading a malicious model repository can trigger execution of attacker-controlled code despite the user's explicit opt-out. The issue is fixed in vLLM 0.18.0.

For your environment

Are you exposed to this one?

Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.

Start free trial

ANALYST BRIEF

Impact, mitigation & remediation

What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.

Impact

What an attacker gets, and what they’ve been doing with it.

Successful exploitation can result in remote code execution in the context of the vLLM process when a malicious model repository is loaded. Because the vulnerable behavior defeats an explicit security opt-out, affected deployments may execute untrusted code under the false assumption that remote code trust has been disabled. The stated CVSS vector indicates high impact to confidentiality, integrity, and availability.

Mitigation

If you can’t patch tonight, do this now.

Until upgrading, avoid loading untrusted or third-party model repositories, especially any repository that may rely on remote custom code. Restrict model sources to vetted, internally controlled, or cryptographically verified repositories where possible. Run vLLM in a tightly sandboxed environment with minimal privileges and strong isolation to reduce the blast radius of unintended code execution. Monitor and review model-loading workflows for any use of repositories that could supply attacker-controlled code.

Remediation

Patch, then assume compromise.

Upgrade vLLM to version 0.18.0 or later, which patches the hardcoded trust_remote_code=True behavior. The provided references indicate the fix is associated with commit 00bd08edeee5dd4d4c13277c0114a464011acf72 and pull request #36192. Validate after upgrade that model-loading paths correctly honor --trust-remote-code=False.

PUBLIC EXPLOITS

Exploits

No public exploits tracked yet. Mallory keeps watching.

VALID 0 / 0 TOTALView more in app

No public exploit code observed for this vulnerability.

EXPOSURE SURFACE

Affected products & vendors

Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.

VendorProductType

VllmVllmapplication

Vllm-ProjectVllmapplication

Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.

ACTIVITY FEED

Recent activity

5 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.

5 SOURCESView more in app

cvefeed high severityNews

Mar 27, 2026

CVE-2026-27893 - vLLM's hardcoded trust_remote_code=True in NemotronVL and KimiK25 bypasses user security opt-out

A remote code execution vulnerability in vLLM caused by hardcoded `trust_remote_code=True` in two model implementation files, which bypasses the user's explicit security setting and allows malicious model repositories to execute code.

What this page doesn’t show

The version that knows your environment.

This page is what’s public. Mallory adds the parts that aren’t: which of your assets are affected, which adversaries are exploiting it right now, which detections to deploy, and what to do tonight.

Start free trial

Exposure mapping

Query your assets running an affected version, and investigate the blast radius.

Threat actor evidence

Every observed campaign linking this CVE to a named adversary.

Associated malware

Malware families riding this exploit, with evidence and IOCs.

Detection signatures

YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.

Vendor-by-vendor mapping

Cross-references every affected SKU, including bundled OEM variants.

Social activity4

Community discussion across Reddit, Mastodon, and other social sources.

EXTERNAL REFERENCES

MITRE CVE

cve.org · CVE-2026-27893

NVD

nvd.nist.gov/vuln/detail/CVE-2026-27893

MITRE CWE · CWE-693

Protection Mechanism Failure

Patch

github.com/vllm-project/vllm/commit/00bd08edeee5dd4d4c13277c0114a464011acf72

Vendor Advisory

github.com/vllm-project/vllm/security/advisories/GHSA-7972-pg2x-xr59

Issue Tracking

github.com/vllm-project/vllm/pull/36192