Agentic Code Scanning - EP 54 - Rome Thorstenson - Rafter.so

In this episode, Philippe Trounev interviews Rome Thorstenson, a software engineer and AI researcher, discussing the intersection of AI and cybersecurity. They explore the current state of code security, the role of AI agents in identifying vulnerabilities, and the challenges of trusting these systems. Rom shares insights from his research at NeurIPS and emphasizes the importance of proactive security measures for developers.takeaways80% of the code shipped to production is not secure.AI agents are increasingly used to analyze code for vulnerabilities.Security often takes a backseat to feature development.Evaluating the security of a code base is a complex task.Prompt injection poses significant risks for AI systems.Developers need to prioritize security in their workflows.Rafter offers tools to simplify security scanning for developers.Research in mechanistic interpretability can enhance AI security agents.The landscape of cybersecurity is evolving with AI advancements.Proactive security measures are essential to combat emerging threats.titlesAI's Role in Cybersecurity: A Deep DiveUnderstanding Code Vulnerabilities with AI AgentsSound Bites"AI writes most of the code.""80% of the code is not secure.""Prompt injection is a huge problem."Chapters00:00Introduction to AI Agents in Cybersecurity02:41The State of Code Security and Vulnerabilities05:10Building AI Agents for Code Analysis07:52Evaluating AI Agents and Benchmarking10:27Autonomous Feedback Loops in Cybersecurity13:07Trusting AI Agents for Security Fixes15:47Understanding Vulnerabilities and AI's Role18:42Real-World Examples of Vulnerability Detection23:25Navigating App Development Challenges24:32Getting Started with Rafter28:03Understanding Mechanistic Interoperability35:06Interpreting Model Features and Security37:49Top Security Practices for Developershttps://www.docsie.ioJoin us on Discord https://discord.gg/pAUGNTzv