Description
LightLLM through 1.2.0 visual_only deployments expose an unauthenticated RPyC service with allow_pickle enabled that deserializes attacker-supplied arguments in the remote_infer_images method. Attackers can reach the visual RPyC port and pass objects with __reduce__ methods to execute arbitrary code with service account privileges.
Published: 2026-09-30
Score: 9.3 Critical
EPSS: < 1% Very Low
KEV: No
Impact: Remote Code Execution
Action: Immediate Patch
AI Analysis

Impact

LightLLM deployments that use the visual_only feature expose an unsecured Remote Procedure Call (RPyC) service. The service runs with the application's privileges and has allow_pickle enabled, meaning it accepts pickled Python objects from anyone who can reach the visual-only port. An attacker can craft a pickle payload containing a __reduce__ method, which the RPyC service will unpickle and execute, allowing arbitrary code execution. This flaw is formally classified as CWE‑502, injection of untrusted data via serialization.

Affected Systems

The vulnerability affects the ModelTC:LightLLM product through version 1.2.0. Any deployment that has not been upgraded beyond 1.2.0 and that exposes the visual-only RPyC port is potentially vulnerable. Users running the open‑source implementation hosted on GitHub or the referenced repositories should verify the version they use.

Risk and Exploitability

The severity of the flaw is high, with a CVSS score of 9.3. The EPSS score is not available, and the vulnerability is not listed in CISA KEV, suggesting it has not yet been widely exploited. Attackers can reach the vulnerable service without authentication, exploiting the unguarded RPyC port. Since the service account privileges run with the application, code execution would occur with full privilege of that account, giving the attacker control over the system.

Generated by OpenCVE AI on September 30, 2026 at 17:37 UTC.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Upgrade LightLLM to a version newer than 1.2.0 or apply the vendor patch that disables allow_pickle in the visual-only service.
  • Restrict access to the visual-only RPyC port using firewalls or VPNs so that only trusted hosts can connect.
  • Reconfigure the service to set allow_pickle=False or remove the visual-only RPyC interface if it is not required.

Generated by OpenCVE AI on September 30, 2026 at 17:37 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories yet.

History

Thu, 01 Oct 2026 04:30:00 +0000

Type Values Removed Values Added
First Time appeared Modeltc
Modeltc lightllm
Vendors & Products Modeltc
Modeltc lightllm

Wed, 30 Sep 2026 20:30:00 +0000

Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'yes', 'Exploitation': 'poc', 'Technical Impact': 'total'}, 'version': '2.0.3'}


Wed, 30 Sep 2026 15:00:00 +0000

Type Values Removed Values Added
Description LightLLM through 1.2.0 visual_only deployments expose an unauthenticated RPyC service with allow_pickle enabled that deserializes attacker-supplied arguments in the remote_infer_images method. Attackers can reach the visual RPyC port and pass objects with __reduce__ methods to execute arbitrary code with service account privileges.
Title LightLLM through 1.2.0 Unauthenticated Remote Code Execution via Visual-Only RPyC Service
Weaknesses CWE-502
References
Metrics cvssV3_1

{'score': 9.8, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H'}

cvssV4_0

{'score': 9.3, 'vector': 'CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N'}


Subscriptions

Modeltc Lightllm
cve-icon MITRE

Status: PUBLISHED

Assigner: VulnCheck

Published:

Updated: 2026-09-30T19:38:06.768Z

Reserved: 2026-09-30T14:28:17.072Z

Link: CVE-2026-103395

cve-icon Vulnrichment

Updated: 2026-09-30T19:37:52.709Z

cve-icon NVD

Status : Deferred

Published: 2026-09-30T15:22:27.710

Modified: 2026-09-30T20:17:30.840

Link: CVE-2026-103395

cve-icon Redhat

No data.

cve-icon OpenCVE Enrichment

Updated: 2026-10-01T04:15:16Z

Weaknesses
  • CWE-502

    Deserialization of Untrusted Data