CVE-2026-108759
MediumCVSS 6.8Summary
mistral.rs 0.9.0 through 0.9.4 contains a link following vulnerability in mistralrs-code-exec that allows sandboxed shell code to read and overwrite files outside the sandbox via symlinks. Attackers or prompt-injected agents can name symlinks as outputs or reuse sessions with symlinked input paths to access files with the server process's permissions.
Risk Assessment
The risk includes unauthorized access to files outside the sandbox, which can lead to data confidentiality and integrity breaches and privilege escalation.
Recommendation
Update mistral.rs to a patched version. Additionally, restrict server process permissions and enforce strict symlink policies.
Other vulnerabilities in mistral.rs
See all- CVE-2026-108940Medium
A vulnerability in EricLBuehler Mistral.rs up to 0.9.3 affects the function calibration_apply in mistralrs-server-core/src/handlers.rs. Manipulation of the argument save_cimatrix results in path traversal. The attack can be initiated remotely. The exploit is public. Upgrading to 0.9.4 fixes the issue.
- CVE-2026-108939Medium
A vulnerability in EricLBuehler mistral.rs up to 0.9.3 affects the function is_global_ip in mistralrs-server-core/src/media_source.rs. Manipulation of the argument ip leads to server-side request forgery. The attack can be launched remotely. The exploit is public. Upgrading to 0.9.4 is recommended.
- CVE-2026-75090Medium
A vulnerability was detected in EricLBuehler Mistral.rs up to 0.8.22, affecting the function convert_gguf_to_hf_tokenizer of the file mistralrs-core/src/gguf/gguf_tokenizer.rs in the GGUF Tokenizer component. Manipulation of the arguments eos_token_id/bos_token_id/unknown_token_id results in out-of-bounds read. The attack can be executed remotely and the exploit is now public. Upgrading to version 0.8.23 can resolve this issue.
Original NVD description (English source)
mistral.rs 0.9.0 through 0.9.4 contains a link following vulnerability in mistralrs-code-exec that allows sandboxed shell code to read and overwrite files outside the sandbox via symlinks. Attackers or prompt-injected agents can name symlinks as outputs or reuse sessions with symlinked input paths to access files with the server process's permissions.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS