Hidden Risks in Open-Source Code and AI Models
summary
Shows that popular open-source packages and pretrained models get weaponized: a published package-lab demo creates a fake pip package that autoloads remote code; a honeypot VM records an attacker gaining interactive shell and exfiltrating secrets; a npm package brings 800+ transitive deps and attackers perform account takeovers to push miners to packages with 20M+ weekly downloads; malicious Hugging Face models exploit pickle-based model backdoor (author reports ~80% of models use pickle) — remote shells and stealthy exfiltration work in practice.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 2:50What is supply chain security
- 5:31Fighting attackers
- 7:15Transitive dependencies
- 9:35Case study: UA parser JS
- 13:56Case study: node-ipc
- 17:03Typosquatting in Python
- 20:04Demonstrating the attack
- 23:17Clipboard hijacking
- 26:24Remote shell infection
- 30:57Large-scale campaigns
- 35:10Social engineering scams
- 38:16AI in software supply chain
- 44:00Pickle serialization risks
- 47:31Model poisoning
- 50:26Best practices and QA
related talks
36mThe Dark Side of AI: The Hidden Risks in Open-Source AI Models
45mThe Dark Side of AI: The Hidden Supply Chain Risks in Open...
30mThe Simple, Yet Lethal, Anatomy of an AI Attack
38mThe Risks of Blind Trust in Code from Strangers
37mBeyond the Basics: Hacking and Securing Machine Learning Systems and Environments
28mOwn Goal: How AI Agents Are Quietly Scoring for the Other Side
32mFuture Open Source LLM Kill Chains
46mInsane in the Supply Chain: Threat modeling for...