New critical vulnerabilities reported in Open Source AI/ML tools

Protect AI reports critical vulnerabilities discovered in various open-source AI/ML. The AI security firm detailed a total of 32 security defects reported through its Huntr AI bug bounty program.

Critical-severity issues

• CVE-2024-22476 (CVSS score 10): SQL Injection and RCE in Intel Neural Compressor software. Recommendations: Upgrade to the latest release.

• CVE-2024-3234 (CVSS score 9.8): LFI due to the use of outdated components in ChuanhuChatGPT. Recommendations: Upgrade to version 20240305.

• CVE-2024-3429 (CVSS score 9.8): Arbitrary file reading via path traversal in LoLLMs. Recommendations: Upgrade to version 9.6.

• CVE-2024-3584 (CVSS score 9.8): Path traversal in collection name leads to arbitrary file overwrite in Qdrant. Recommendations: Upgrade to version v1.9.0.

• CVE-2024-3829 (CVSS score 9.8): Arbitrary file read and write during snapshot recovery in Qdrant. Recommendations: Upgrade to version v1.9.0.

• CVE-2024-4146 (CVSS score 9.8): User can access unauthorized projects from org in Lunary. Recommendations: Upgrade to version 1.2.26.

• CVE-2024-3149 (CVSS score 9.6): SSRF in the upload link feature leads to accessing internal Collector API and escalating attack to arbitrary file deletion and limited LFI in AnythingLLM. Recommendations: Upgrade to the latest release.

• CVE-2024-5128 (CVSS score 9.4): IDOR allowing view/update/delete of any dataset_prompt/dataset_prompt_variation in any dataset/projects in Lunary. Recommendations: Upgrade to version 1.2.25.

• CVE-2024-3761 (CVSS score 9.1): Missing authorization on delete datasets in Lunary. Recommendations: Upgrade to version 1.2.8.

• CVE-2024-4315 (CVSS score 9.1): Lack of path sanitization for Windows leads to LFI in LoLLMs. Recommendations: Upgrade to version 9.8.

• CVE-2024-5211 (CVSS score 9.1): Path traversal to arbitrary file read/delete/overwrite, DoS attack, and admin account takeover in AnythingLLM. Recommendations: Upgrade to the latest release.

• CVE-2024-0087 (CVSS score 9.0): Arbitrary file creation/appending in log file configuration interface can lead to remote code execution in Nvidia Triton Inference server. Recommendations: Upgrade to version 24.04.

Additionally, the report identified a dozen high-severity vulnerabilities in LoLLMs, Lunary, AnythingLLM, Deep Java Library (DJL), Scrapy, and Gradio.

Protect AI notes that all vulnerabilities were reported to the respective maintainers at least 45 days before the publication of the report.