CVE-2025-62608
📋 TL;DR
CVE-2025-62608 is a heap buffer overflow vulnerability in MLX's load() function when parsing malicious NumPy .npy files. Attackers can trigger a 13-byte out-of-bounds read, potentially causing crashes or information disclosure. This affects all users of MLX versions before 0.29.4 who process untrusted .npy files.
💻 Affected Systems
- MLX (array framework for machine learning on Apple silicon)
📦 What is this software?
Mlx by Ml Explore
⚠️ Risk & Real-World Impact
Worst Case
Remote code execution leading to complete system compromise if the out-of-bounds read can be weaponized into arbitrary code execution.
Likely Case
Application crash (denial of service) or limited information disclosure from heap memory.
If Mitigated
No impact if patched version is used or untrusted .npy files are not processed.
🎯 Exploit Status
Exploitation requires the attacker to supply a malicious .npy file that gets processed by the vulnerable load() function.
🛠️ Fix & Mitigation
✅ Official Fix
Patch Version: 0.29.4
Vendor Advisory: https://github.com/ml-explore/mlx/security/advisories/GHSA-w6vg-jg77-2qg6
Restart Required: No
Instructions:
1. Update MLX to version 0.29.4 or later using pip: 'pip install --upgrade mlx>=0.29.4' 2. Verify the update completed successfully. 3. No restart required for Python applications, but restart any long-running MLX processes.
🔧 Temporary Workarounds
Input validation for .npy files
allImplement strict validation of .npy files before passing to mlx::core::load()
Sandbox MLX processing
allRun MLX in isolated containers or sandboxes when processing untrusted files
🧯 If You Can't Patch
- Implement strict access controls to prevent untrusted users from supplying .npy files
- Monitor for crashes in MLX processes and investigate any .npy file processing failures
🔍 How to Verify
Check if Vulnerable:
Check MLX version: 'python -c "import mlx; print(mlx.__version__)"' - if version < 0.29.4, system is vulnerable.Check Version:
python -c "import mlx; print(mlx.__version__)"Verify Fix Applied:
After updating, verify version is 0.29.4 or higher using same command. Test with known safe .npy files to ensure functionality.📡 Detection & Monitoring
Log Indicators:
- Segmentation faults or crashes in MLX processes
- Error messages related to buffer overflows or memory corruption in MLX logs
- Unexpected termination of MLX file processing jobs
Network Indicators:
- Unusual file uploads to MLX processing endpoints
- Multiple failed .npy file processing attempts
SIEM Query:
source="mlx_logs" AND ("segmentation fault" OR "buffer overflow" OR "memory corruption")