CVE-2025-66455
9.8 CRITICALpublic exploit availablePublished 2026-09-18 · Updated 2026-09-23
AI risk analysis
- Summary
- LMDeploy's PyTorch DistServe/PD-disaggregation control plane deserializes messages using Python pickle, allowing remote code execution if an attacker can reach the `/distserve/p2p_connect` endpoint.
- Exploitability
- Exploitation is relatively easy if an attacker can reach the `/distserve/p2p_connect` endpoint, which requires the DistServe API server to be exposed to untrusted clients.
- Blast radius
- If exploited, this could result in unauthenticated remote code execution with the privileges of the LMDeploy serving process, potentially leading to full system compromise.
- Detection
- No reliable host or network indicator is derivable from the published description.
- Prioritized remediation
- Restrict access to the `/distserve/*` endpoints to trusted networks and enable API-key authentication. Upgrade to LMDeploy 0.16.0 or later.
Analysis generated locally by qwen2.5:7b-instruct (no data left the box). AI-assisted — verify against primary sources before acting.
NVD description
LMDeploy is a toolkit for compressing, deploying, and serving large language models. Starting in version 0.9.2 and prior to version 0.16.0, LMDeploy's PyTorch DistServe/PD-disaggregation control plane used `recv_pyobj()` to deserialize messages received through a ZeroMQ PULL socket. PyZMQ implements `recv_pyobj()` using Python pickle deserialization, which can execute arbitrary code while reconstructing an object. The peer address used by the receiver was supplied through the `POST /distserve/p2p_connect` HTTP endpoint. An attacker who could reach an affected DistServe API server could cause the server to connect to an attacker-controlled ZeroMQ endpoint and deserialize a crafted pickle payload. API-key authentication is not enabled unless the operator explicitly configures it. As a result, affected DistServe deployments without API keys allowed unauthenticated remote code execution with the privileges of the LMDeploy serving process. This issue affects the PyTorch backend when PD-disaggregation/DistServe is enabled. Ordinary deployments that do not use the affected disaggregated-serving path do not expose this data flow. The fix was released in LMDeploy 0.16.0. Users who cannot upgrade immediately should prevent untrusted clients from reaching `/distserve/*` endpoints, restrict the DistServe HTTP and ZeroMQ control planes to trusted cluster networks, configure API-key authentication, and block arbitrary outbound ZeroMQ connections from serving nodes. These measures reduce exposure but do not make pickle deserialization safe.
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
- HIGHCVE-2026-67615PoC
- CRITICALCVE-2026-94301
- HIGHCVE-2025-51457
- CRITICALCVE-2026-100291
- HIGHCVE-2026-100520PoC
- HIGHCVE-2026-100552PoC
- CRITICALCVE-2026-100716PoC
- CRITICALCVE-2026-100717PoC
Related by shared AI tags and CWE weakness class. Browse the full archive.