Investigate in your agent.

Turn this advisory into a focused investigation. Ask your agent to check your project’s versions, explain the evidence, and suggest a compatible fix.

Your agent needs remote MCP support and access to https://cve.new/mcp.

Read the setup guide

Skip to content
Check your stack
CVE-2026-115915

vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCE

Products / packages vllm-project vllm · vllm

Published

Share

Severity and exploitation

Not in CISA KEVSource: HIGH

KEV coverage does not rule out exploitation.

KEV checked 15/11/2026, 02:40:44 UTC
GitHub_M · CVSS:3.1 · 8.1
Assessment vectorCVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H
OSV · CVSS:3.1 · 8.1
Assessment vectorCVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H
OSV · CVSS:3.1 · 9.8
Assessment vectorCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
The headline uses the highest imported catalog score. Assessments can differ by source and CVSS version.

Severity

critical9.8

CVSS measures technical severity, not the likelihood of an attack.

Known exploitation

Not listed in the checked CISA KEV catalog

Absence from CISA KEV does not rule out attacks.

Public exploit

Not established by this record

Missing evidence does not mean no public exploit exists.

Remediation

Package fixes recorded

Fix versions are recorded for 1 of 1 mapped packages. Check each release branch and any separate product guidance.

What this means

vLLM 0.30.0 is vulnerable to denial of service or Ghostscript RCE when processing untrusted image bytes, specifically EPS inputs, due to improper input validation. An attacker with access to the image-processing endpoint can trigger command execution as the vLLM process user if the host has a vulnerable Ghostscript build. A separate denial-of-service condition exists with Pillow 12.0.0-12.2.0.

Summarized by @cf/meta/llama-3.3-70b-instruct-fp8-fast. Verify details against the original sources.

Attack conditions

Exploitation requires an image-capable request path, a caller able to reach it under the deployment's authentication policy, and a vulnerable host Ghostscript build. The published vllm==0.30.0 requirements leave Pillow unpinned, so an existing environment can retain an affected Pillow version; the source at commit bc21cba9673cfc2256a2726b4bc32f062237cd35 requires pillow>=12.3.0.

Excerpt from the original source description below.

Review supporting sources →

Affected versions

Check the exact model, release, and any platform requirements against your installation.

vllm-project · GitHub_M

vllm

1 range

Compare with this publisher’s ranges. This does not inspect your installation.

affected

< 0.32.0

unaffected

unknown

Other versions: unknown

Packages

Compare each installed version with the publisher’s affected ranges. Where a fix is recorded, choose the appropriate release branch and test compatibility before upgrading.

vllm PyPI
Affected range Recorded fix
Earlier releases before 0.32.0Ecosystem versions0.32.0
From 0.30.0 before 0.32.0Ecosystem versions0.32.0

Version evidence: OSV · GHSA-h3rc-j82p-2c3hOSV · PYSEC-2026-4346

View original version evidence (101 explicit versions)

What to do

Compare your installed product and version with the affected versions above. A matching product name alone does not confirm exposure.

Use the recorded fix for your release branch. Review the publisher’s instructions and test compatibility before upgrading.

cve.new guidance unless attributed to a publisher. This page does not inspect your installation.

Public exploit evidence

No public exploit reference is identified in the imported evidence. This is not a claim that no exploit exists.

Original source description

Read the full publisher description

Summary

In vLLM 0.30.0, ImageMediaIO passes attacker-controlled image bytes to Pillow's format autodetection without restricting the accepted formats; the request's MIME label does not constrain decoder selection. An EPS payload labeled image/png reached the host's Ghostscript executable. On a host with Ghostscript affected by CVE-2024-29510, this exposes the known Ghostscript sandbox escape through vLLM's image-loading path and can execute commands with the vLLM process's privileges. A bounded marker-command payload executed through the PyPI wheel's actual ImageMediaIO.load_base64() implementation using Debian Ghostscript 10.03.0. The loader-level execution is reproduced; the HTTP request path is source-traced but was not exercised end to end. This report concerns vLLM's untrusted-format dispatch and does not claim a new Ghostscript vulnerability. Exploitation requires an image-capable request path, a caller able to reach it under the deployment's authentication policy, and a vulnerable host Ghostscript build.

Details

ImageMediaIO is the boundary where attacker-controlled bytes enter Pillow's format autodetection: the declared MIME type does not constrain decoder selection, and the loader calls Image.open(BytesIO(data)) without a Pillow format allowlist. Pillow recognizes EPS by its file signature, and vLLM's subsequent image.load() call enters Pillow's EPS processing path. The OpenAI chat multimodal parser reaches this path through parse_image() and fetch_image() and the media connector; the data URL path extracts the MIME label and forwards the bytes to the loader (source). This is the vLLM-specific issue: untrusted image bytes can select a Pillow-supported format parser. Each downstream impact depends on the relevant parser, component version, and host configuration.

Pillow recognizes EPS by its file signature and invokes Ghostscript with -dSAFER, -sDEVICE=pnmraw, and the untrusted EPS input (Pillow EPS plugin, load path). The command execution in this report uses the already-known Ghostscript CVE-2024-29510, which allows crafted PostScript to bypass -dSAFER; Ghostscript lists 10.03.1 as the upstream fixed version (Ghostscript CVE details). This is an N-day dependency chain: CVE-2024-29510 provides the command-execution primitive, while the vLLM-specific issue is that untrusted, remotely supplied image bytes can reach the affected interpreter. No new Ghostscript flaw is claimed here.

The vLLM Python requirements declare Pillow but not Ghostscript, so this execution chain depends on a Ghostscript executable supplied by the host or deployment image. I reproduced command execution with the PyPI vllm==0.30.0 wheel, Pillow 12.3.0, and Debian amd64 Ghostscript 10.03.0~dfsg-1. The Codean Labs CVE-2024-29510 testkit identified that Ghostscript build as vulnerable. The adapted EPS payload ran only touch on a temporary marker. The marker existed after the loader call, although Pillow then raised ValueError because the exploit changes Ghostscript's output device. The API route was traced in source but a model-backed HTTP request was not run. Only PyPI vllm==0.30.0 was checked; no wider affected release range or other Ghostscript vendor build is claimed.

The same input boundary also exposes a separate Pillow-only denial-of-service condition when the installed Pillow version is 12.0.0–12.2.0. CVE-2026-59203 is triggered by a negative %%BeginBinary byte count and causes Pillow's EPS parser to loop during Image.open(); it does not invoke Ghostscript or provide code execution (Pillow advisory). The published vllm==0.30.0 requirements leave Pillow unpinned, so an existing environment can retain an affected Pillow version; the source at commit bc21cba9673cfc2256a2726b4bc32f062237cd35 requires pillow>=12.3.0. In an isolated local check with Pillow 12.2.0, the official sample hung through ImageMediaIO.load_bytes() and MediaConnector.fetch_image_async(). Eight concurrent image loads occupied the connector's eight media-loading threads and left a benign image load queued. Pillow 12.3.0 rejected the sample promptly. This confirms a conditional media-processing DoS path, separate from the Ghostscript RCE chain; full HTTP-server availability was not measured.

Impact

When a vLLM image-processing endpoint accepts attacker-supplied content and the host has Ghostscript affected by CVE-2024-29510, a caller who can reach the endpoint can trigger command execution as the vLLM process user. API-key authentication and network access controls determine who can reach the request path. Deployments without Ghostscript or with a build fixed for CVE-2024-29510 are not affected by this execution chain. Separately, existing environments that run vLLM 0.30.0 with Pillow 12.0.0–12.2.0 are exposed to CVE-2026-59203 through the same image-format dispatch boundary; the demonstrated impact is exhaustion of media-loading workers, and this Pillow CVE does not cause RCE. The report asks vLLM to assess and address the unsafe dispatch of untrusted image bytes and the affected dependency combinations.

This advisory was fixed in MR60022.

References

6 sources

Inside the record

Evidence available

Package identity
Version evidence
Remediation guidance · not supplied
Weakness classification
Additional publishers

Describes the evidence in this view, not whether the product is safe.

Original records

CVE-2026-115915

View the latest unmodified CVE JSON from the official source. It may be newer than this imported page.

cve.new JSON →

Underlying weaknesses

The source’s classification of what caused the vulnerability.

CWE-20 ↗

CWE-20: Improper Input Validation

GitHub_M

Related →
CWE-183 ↗

CWE-183: Permissive List of Allowed Inputs

GitHub_M

Related →
Contributing organizations · 2
GitHub_M

Assigning organization

2026-10-26
CISA-ADP

Additional data publisher

2026-10-26

Timeline

All times UTC
  1. Advisory published

    Publication date supplied by the source.

  2. Source updated

    Latest modification date supplied by the source.

  3. Last imported by cve.new

    Latest catalog import, not the date we first saw this advisory.

What changed

Follow changes
No substantive changes have been recorded here yet. History starts when cve.new observes a change after tracking is enabled; earlier edits are not reconstructed.

Large field values are marked as truncated previews. Original source snapshots remain retained.