Remote Code Execution in vLLM Affected by Unsafe Deserialization Using Mooncake
CVE-2025-29783

10CRITICAL

Key Information:

Status
Vendor
CVE Published:
19 March 2025

What is CVE-2025-29783?

CVE-2025-29783 is a critical vulnerability identified in the vLLM, a high-throughput and memory-efficient inference and serving engine for large language models (LLMs). The vulnerability arises when vLLM is configured to utilize Mooncake, leading to unsafe deserialization vulnerabilities that can be exploited through ZMQ/TCP connections across all network interfaces. This presents significant risks to organizations utilizing vLLM for their distributed applications, potentially allowing malicious actors to remotely execute code on affected systems, which can lead to data breaches or system manipulations.

Technical Details

The vulnerability in CVE-2025-29783 lies in the way vLLM handles data deserialization when integrated with Mooncake. Specifically, the insecure handling of deserialization over network protocols makes it possible for attackers to craft and send malicious payloads that a vulnerable vLLM instance would improperly execute. This creates a severe threat, as it could allow attackers access to execute arbitrary code on remote hosts that are part of a distributed setup leveraging vLLM. The issue has been addressed in the software version 0.8.0, which mitigates the vulnerability.

Potential Impact of CVE-2025-29783

  1. Remote Code Execution: The primary impact of CVE-2025-29783 is the potential for remote code execution, which allows attackers to gain control over affected systems. This could enable further exploits, including data manipulation, unauthorized data access, or the introduction of malware.

  2. Data Breaches: Organizations may experience significant data security incidents as a result of unauthorized access enabled by this vulnerability. Attackers could exfiltrate sensitive information stored on compromised systems, leading to potential regulatory and compliance violations.

  3. Operational Disruption: Exploiting this vulnerability could lead to service interruptions or system downtime, resulting in operational challenges for organizations relying on vLLM for mission-critical applications. Such disruptions can result in financial losses and damage to reputation.

Affected Version(s)

vllm >= 0.6.5, < 0.8.0

References

CVSS V3.1

Score:
10
Severity:
CRITICAL
Confidentiality:
High
Integrity:
High
Availability:
High
Attack Vector:
Network
Attack Complexity:
Low
Privileges Required:
None
User Interaction:
None
Scope:
Changed

Timeline

  • Vulnerability published

  • Vulnerability Reserved

.
CVE-2025-29783 : Remote Code Execution in vLLM Affected by Unsafe Deserialization Using Mooncake