← Back to search

CVE-2026-96560

9.8 CRITICALpublic exploit available

Published 2026-09-23 · Updated 2026-09-23

AI risk analysis

Summary
LightLLM through 1.2.0 has a remote code execution vulnerability due to an unauthenticated RPyC control channel that deserializes attacker-supplied data, allowing arbitrary code execution with the service account privileges.
Exploitability
Exploitation is relatively straightforward given the unauthenticated nature of the RPyC control channel, and requires the --pd_trans_mode nccl flag to be set during service startup.
Blast radius
If exploited, the vulnerability could lead to complete compromise of the LightLLM service, potentially allowing attackers to execute arbitrary code with the service account privileges.
Detection
No reliable host or network indicator is derivable from the published description.
Prioritized remediation
Disable the RPyC control channel or ensure that the --pd_trans_mode flag is not set to nccl during service startup. If disabling the feature is not feasible, upgrade to a version that addresses this vulnerability, such as LightLLM 1.2.1 or later.
rceauth-bypassremoteservice

Analysis generated locally by qwen2.5:7b-instruct (no data left the box). AI-assisted — verify against primary sources before acting.

NVD description

LightLLM through 1.2.0 contains a remote code execution vulnerability in the KV-transfer worker when started with --pd_trans_mode nccl, which exposes an unauthenticated RPyC control channel that deserializes attacker-supplied data. Attackers can send malicious pickled objects to the exposed RPyC ThreadedServer to execute arbitrary code with the privileges of the LightLLM service account.

CVSS vector

CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H

Weaknesses

CWE-502

Public exploit & PoC references

All references

Source data: NVD (nvd.nist.gov), public domain. Exploit-DB.ai adds local AI analysis for defensive use only.

Related CVEs

Related by shared AI tags and CWE weakness class. Browse the full archive.