Picklescan Bugs Allow Malicious PyTorch Models to Evade Scans and Execute Code
Summary: JFrog researchers disclosed three critical flaws in Picklescan, an open-source scanner maintained by Matthieu Maitre that parses Python pickle files to detect suspicious imports before execution, allowing attackers to bypass its blocklist and execute arbitrary code through malicious PyTorch models. The flaws were addressed in Picklescan version 0.0.31, released September 9 following responsible disclosure on June 29, 2025, and come alongside a separate high-severity bypass, CVE-2025-46417 (CVSS 7.5/7.1), detailed by SecDim and DCODX.
Key Takeaway: Picklescan works from a blocklist, so a payload using modules not on the deny-list, such as linecache and ssl, returned 'no issues found' in version 0.0.24 while capable of reading /etc/passwd and exfiltrating it through DNS logs. Organizations relying on a single scanning tool to vet ML models are exposed to exactly the supply chain attack these bypasses enable.
Read the original at The Hacker News ↗
- Topics
- Supply Chain & AppSec
- Mentions
- CVE-2025-46417
- Related
- Kitesurf is now open source Cyber Risks · Nov 13, 2026Trained to betray: How post-training supply chain attacks hijack coding agents Cyber Risks · Nov 12, 2026ZDI-26-913: (0Day) Meta fairseq load_audio_path Code Injection Remote Code Execution Vulnerability Cyber Risks · Nov 11, 2026