CVE-2026-78147

Received Received - Intake

Deserialization Flaw in llama.cpp RPC Server

Publication date: 2026-08-23

Last updated on: 2026-08-23

Assigner: VulDB

Description

A vulnerability was found in ggml-org llama.cpp bec4772f6. The impacted element is the function deserialize_tensor of the file ggml/src/ggml-rpc/ggml-rpc.cpp of the component ggml-RPC Server. Performing a manipulation of the argument op/op_params results in deserialization. The attack may be initiated remotely. This vulnerability is distinct from CVE-2026-34159 (GHSA-j8rj-fmpv-wcxw, PR #20908), which only added a buffer==nullptr rejection in create_node() and does not validate op or op_params. The reported GitHub issue was closed automatically due to inactivity.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published

2026-08-23

Last Modified

2026-08-23

Generated

2026-08-24

AI Q&A

2026-08-24

EPSS Evaluated

N/A

NVD

CVE-2026-78147

EUVD

EUVD-2026-64671

Affected Vendors & Products

Vendor	Product	Version / Range
ggml-org	llama.cpp	*

Helpful Resources

Exploitability

CWE

KEV

CWE ID	Description
CWE-502	The product deserializes untrusted data without sufficiently ensuring that the resulting data will be valid.
CWE-20	The product receives input or data, but it does not validate or incorrectly validates that the input has the properties that are required to process the data safely and correctly.

Attack-Flow Graph

Executive Summary

This vulnerability exists in the ggml-org llama.cpp library, specifically in the ggml-RPC Server component. It involves a flaw in the deserialize_tensor function where manipulating the op/op_params argument can lead to deserialization issues. The attack can be initiated remotely without requiring authentication or user interaction.

Detection Guidance

This vulnerability involves deserialization issues in the ggml-RPC Server component of llama.cpp. Detection may require inspecting network traffic for RPC calls to the vulnerable function deserialize_tensor, particularly those manipulating op/op_params arguments. Check for unexpected deserialization attempts or crashes in the ggml-rpc.cpp file.

Impact Analysis

This vulnerability could allow remote attackers to execute arbitrary code or cause denial of service by exploiting the deserialization flaw. It may lead to unauthorized access, data leakage, or system compromise depending on the application's use of the affected component.

Compliance Impact

The vulnerability involves deserialization of manipulated arguments in the ggml-RPC Server, which could lead to remote code execution or unauthorized data access. This may impact compliance with GDPR by risking unauthorized processing of personal data and HIPAA by potentially exposing protected health information.

Mitigation Strategies

Update to the latest version of ggml-org llama.cpp to address the deserialization issue in the ggml-RPC Server component. Disable the ggml-RPC Server if not in use to prevent remote exploitation.

Hi! I’m here to help you understand CVE-2026-78147. Ask me anything about the vulnerability, its impact, or mitigation strategies.

0/70