Secure AI Orchestration: Mitigating Model-centric Attacks with Flyte
Learn about model-centric security attacks in machine learning and their mitigation strategies in this technical talk from OpenSSF. Explore common security vulnerabilities in AI/ML applications, focusing specifically on malicious code injection in pickled model files and potentially harmful code generated by Large Language Models (LLMs). Discover how Flyte, an open-source ML orchestrator, helps protect against these attack vectors through practical demonstrations and examples. Examine the limitations of Flyte's security measures and understand how to enhance protection using additional open-source tools like safetensors and ONNX. Gain valuable insights into securing AI orchestration workflows as machine learning applications continue to expand across text, image, audio, and video generation domains.