The Dark Side of AI: Hidden Supply Chain Risks in Open-Source AI Models

Explore the hidden security risks in open-source AI models through this 45-minute conference talk by Jossef Harush Kadouri and Tzachi Zornshtain from Checkmarx. Gain insights into the construction of open-source AI models, focusing on their vulnerabilities. Witness a live demonstration of exploiting these weaknesses to create malicious models and learn about the ease of publishing them on platforms like HuggingFace. Discover techniques such as embedding malicious code in models and data poisoning for existing models, including targeted attacks on specific victims. Understand the potential dangers lurking in the rapidly growing field of AI tools and the importance of addressing these security concerns in the open-source AI ecosystem.