CVE-2025-66455
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.
Leer descripción completaMostrar menos
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
- Versión: 3.1
- Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
- Puntuación base: 9.8
Probabilidad de explotación (EPSS)
- Probabilidad de explotación en los próximos 30 días: 0.69%
- Percentil entre todas las CVEs puntuadas: 51
- Fecha de la puntuación: 5/10/2026
EPSS (Exploit Prediction Scoring System, de FIRST) estima la probabilidad de que una vulnerabilidad sea explotada en 30 días. Complementa a CVSS (impacto) y a CISA KEV (explotación confirmada).
🎯 Técnicas ATT&CK
Cómo se explota esta vulnerabilidad y qué consigue el atacante, en el lenguaje de MITRE ATT&CK.
- Explotación
T1190Exploit Public-Facing Applicationinitial access95 % - Impacto principal
T1059Command and Scripting Interpreterexecution95 % - Impacto secundario
T1078Valid Accountsstealth · persistence · privilege escalation · initial access60 %
Vector AV:N/AC:L/PR:N/UI:N indica explotación remota sin autenticación (T1190). La deserialization de pickle en recv_pyobj() permite ejecución arbitraria de código (T1059). Sin API keys habilitadas por defecto, el atacante obtiene acceso con privilegios del proceso LMDeploy (T1078).
Inferido por nuestro agente de análisis a partir de la descripción oficial, el vector CVSS y la CWE, y comprobado por un supervisor. Puede contener errores.
🛡️ Mitigaciones ATT&CK que cubren estas técnicas
Tecnologías afectadas (3)
⚠ Inferidas por IA a partir de la descripción — NVD aún no ha analizado esta CVE; no son CPE verificados.
CWE
- CWE-502
Referencias
JSON original (NVD)
Mostrar
{
"id": "CVE-2025-66455",
"cveTags": [],
"metrics": {
"ssvcV203": [
{
"source": "134c704f-9b21-4f2e-91b3-4a467353bcc0",
"ssvcData": {
"id": "CVE-2025-66455",
"role": "CISA Coordinator",
"options": [
{
"exploitation": "none"
},
{
"automatable": "yes"
},
{
"technicalImpact": "total"
}
],
"version": "2.0.3",
"timestamp": "2026-09-18T19:54:08.825899Z"
}
}
],
"cvssMetricV31": [
{
"type": "Secondary",
"source": "security-advisories@github.com",
"cvssData": {
"scope": "UNCHANGED",
"version": "3.1",
"baseScore": 9.8,
"attackVector": "NETWORK",
"baseSeverity": "CRITICAL",
"vectorString": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
"integrityImpact": "HIGH",
"userInteraction": "NONE",
"attackComplexity": "LOW",
"availabilityImpact": "HIGH",
"privilegesRequired": "NONE",
"confidentialityImpact": "HIGH"
},
"impactScore": 5.9,
"exploitabilityScore": 3.9
}
]
},
"affected": [
{
"source": "security-advisories@github.com",
"affectedData": [
{
"vendor": "InternLM",
"product": "lmdeploy",
"versions": [
{
"status": "affected",
"version": ">= 0.9.2, < 0.16.0"
}
]
}
]
}
],
"published": "2026-09-18T18:17:04.420",
"references": [
{
"url": "https://github.com/InternLM/lmdeploy/commit/f05b4ad8bf2e2d84101a1d63b3c44fadd99223b2",
"source": "security-advisories@github.com"
},
{
"url": "https://github.com/InternLM/lmdeploy/releases/tag/v0.16.0",
"source": "security-advisories@github.com"
},
{
"url": "https://github.com/InternLM/lmdeploy/security/advisories/GHSA-2vh9-42vm-xmv2",
"source": "security-advisories@github.com"
}
],
"vulnStatus": "Awaiting Analysis",
"weaknesses": [
{
"type": "Secondary",
"source": "security-advisories@github.com",
"description": [
{
"lang": "en",
"value": "CWE-502"
}
]
}
],
"descriptions": [
{
"lang": "en",
"value": "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."
}
],
"lastModified": "2026-09-23T18:12:04.247",
"sourceIdentifier": "security-advisories@github.com"
}