# Applying DevSecOps Lessons to MLSecOps Source: https://www.youtube.com/watch?v=CLGoVCbudsA Recap page: https://rapidrecap.app/video/CLGoVCbudsA Generated: 2025-09-26T15:36:53.928+00:00 --- ## Quick Overview This panel discussion from Def Con 33 explores the application of DevSecOps principles to MLSecOps, highlighting the challenges and opportunities in securing AI/ML pipelines and the need for cross-disciplinary collaboration. The speakers emphasize the importance of integrating security early in the ML lifecycle and leveraging open-source tools and communities to achieve this. **Key Points:** - DevSecOps principles are increasingly relevant and adaptable to MLSecOps, requiring a shift in mindset towards integrating security throughout the AI/ML lifecycle. - Key challenges in MLSecOps include the complexity of AI/ML systems, the need for new security tooling, and the difficulty in applying traditional security practices to AI models. - Organizations must foster cross-functional collaboration between data scientists, ML engineers, security professionals, and business stakeholders to build secure AI applications. - Open-source tools and communities play a crucial role in developing and disseminating security best practices for AI/ML, providing valuable resources for practitioners. - The OWASP Top 10 for LLM Applications highlights critical threats such as supply chain compromise and data poisoning, emphasizing the need for proactive security measures. - Future maturation of MLSecOps will involve more robust tooling, better understanding of AI-specific vulnerabilities, and a proactive approach to security integrated from the design phase. ![Screenshot at 00:17: The title slide of the panel discussion, 'Applying DevSecOps Lessons to MLSecOps,' featuring the OpenSSF logo and presented at Def Con 33.](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-00-17.png) **Context:** This panel discussion, 'Applying DevSecOps Lessons to MLSecOps,' took place at Def Con 33 and featured experts Christopher Robinson (OpenSSF), Sarah Evans (Dell Technologies), and Eoin Wickens (HiddenLayer). They delved into the critical intersection of AI/ML security and the established practices of DevSecOps. The conversation focused on how to adapt and apply existing security methodologies to the unique challenges presented by machine learning systems and their development pipelines, emphasizing the growing importance of MLSecOps in the current technological landscape. ## Detailed Analysis The panel discussion "Applying DevSecOps Lessons to MLSecOps" from Def Con 33 highlighted the critical need to integrate security principles into the machine learning (ML) lifecycle, drawing parallels with established DevSecOps practices. Speakers Christopher Robinson, Sarah Evans, and Eoin Wickens emphasized that while traditional software engineering security has decades of history, the nascent field of AI/ML security requires adapting these lessons to new, often opaque systems. A key takeaway was the necessity of fostering collaboration between diverse teams—data scientists, ML engineers, security professionals, and business stakeholders—to build secure AI applications. They discussed the challenges of securing ML pipelines, including the unique vulnerabilities of AI models like data poisoning and model evasion, and the lack of mature, specialized tooling compared to traditional software security. The panel also touched upon the OWASP Top 10 ML Security Risks, particularly highlighting supply chain attacks and the difficulty in verifying the integrity of AI models. Open-source tools and communities were identified as vital resources for developing and sharing best practices in MLSecOps, enabling organizations to build more secure and trustworthy AI systems. The speakers projected a future where MLSecOps will become increasingly integrated and automated, requiring continuous learning and adaptation from practitioners. ### Introduction - The panel convenes experts to discuss applying DevSecOps to MLSecOps. ### Challenges in MLSecOps - Complexity of AI/ML systems, evolving threats, and tooling gaps. ### Cross-functional Collaboration - Importance of integrating data scientists, ML engineers, and security professionals. ### Role of Open Source - Leveraging open-source tools and communities for ML security. ### OWASP Top 10 ML Security Risks - Highlighting supply chain attacks and data integrity. ### Future Outlook - Maturation of MLSecOps with better tooling and proactive security. ![Screenshot at 00:00: Title slide for the AIXCC Stage at Def Con 33.](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-00-00.png) ![Screenshot at 00:17: Panel title slide: 'Applying DevSecOps Lessons to MLSecOps', Panel Discussion, Def Con 33, featuring the OpenSSF logo.](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-00-17.png) ![Screenshot at 01:13: Introduction of panelist Christopher "CRob" Robinson, Security Architect at OpenSSF.](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-01-13.png) ![Screenshot at 01:42: Introduction of panelist Sarah Evans, Security Research Program Lead at Dell Technologies.](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-01-42.png) ![Screenshot at 02:03: Introduction of panelist Eoin Wickens, Director of Threat Intelligence at Hidden Layer.](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-02-03.png) ![Screenshot at 02:47: First discussion question: 'How has DevSecOps influenced security integration in new areas like AI/ML?'](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-02-47.png) ![Screenshot at 08:02: Second discussion question: 'What organizational or cultural shifts are needed for successful MLSecOps adoption?'](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-08-02.png) ![Screenshot at 09:51: Third discussion question: 'What new or extended security tooling is most needed for MLSecOps?'](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-09-51.png) ![Screenshot at 13:52: Fourth discussion question: 'What advice would you give practitioners transitioning to AI/ML security from traditional secure development, or vice versa?'](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-13-52.png) ![Screenshot at 18:04: Fifth discussion question: 'Which OWASP ML Security Top 10 threats are most critical to address early, and why?'](https://ss.rapidrecap.app/screens/CLGoVCbudsA/00-18-04.png)