Ratings / AI App

Llama.cpp

Critical   Rating 1 / 5 · Critical — Active surveillance / serious risk.

Claim evaluated

Llama.cpp, despite its local-first design, poses a significant privacy threat due to numerous critical security vulnerabilities. For instance, in April 2024, a remote code execution (RCE) vulnerability (CVSS 9.6) was found in `llama-cpp-python` due to improper handling of chat te

Analysis

Llama.cpp, despite its local-first design, poses a significant privacy threat due to numerous critical security vulnerabilities. For instance, in April 2024, a remote code execution (RCE) vulnerability (CVSS 9.6) was found in `llama-cpp-python` due to improper handling of chat templates, allowing arbitrary code execution. Furthermore, in April 2026, CVE-2026-34159 identified another critical RCE in the RPC backend, enabling unauthenticated attackers to read and write arbitrary process memory. The project's security policy explicitly warns users under "Data privacy" that "To protect sensitive data from potential leaks or unauthorized access, it is crucial to sandbox the model execution," and advises against using the RPC backend in untrusted environments, acknowledging the inherent risks. These vulnerabilities can lead to unauthorized data access and breaches if not properly mitigated by users.

Sources

• https://github.com/ggml-org/llama.cpp/security/policy
• https://www.getsafety.com/blog-posts/security-vulnerability-in-the-llama-cpp-python-package
• https://www.upguard.com/blog/llama-cpp-prompt-leak
• https://www.appsecure.security/vulnerability-database/cve-2026-21869/
• https://www.sentinelone.com/vulnerability-database/cve-2026-34159/