Episode 242 - LLMs Exploiting Vulns, State of DevSecOps

Kyle Kelly joins Seth Law and Ken Johnson as a special guest on the Absolute AppSec podcast. Kyle is an Executive Cybersecurity Consultant at Bancsec, Inc, and Security Researcher at Semgrep, and founder of the wonderful Cramhacks newsletter. As a consultant and researcher, Kyle specializes in supply chain security, a speciality that informs the thoughts he publicizes, but even more so cramhacks reflects his desire to help his readers become contributors to improving the cybersecurity landscape and analysis of software security supply chains. Subscribe to Kyle's newsletter at cramhacks.com.

Bryan Schmidt, information security lead at Adept AI is joining Ken Johnson (@cktricky on twitter/x) and Seth Law (@Sethlaw) for a special episode of Absolute AppSec. Before Adept.AI, Bryan spent the last half decade working as a security engineering manager at, first, Flatiron Health and, later ChowNow, and he worked as a penetration tester and security consultant for that. We’ll be discussing AI during the show as Adept.ai is recently again designated as one of the AI Fortune50. Be sure to tune in to learn a little about Bryan and his trajectory into security and emerging technologies.

Seth and Ken return with analysis of recent research that shows LLMs exploiting known CVEs. And no, it's not completely autonomous yet. This is followed by a breakdown of DataDog's State of DevSecOps article, backing up our gut feel of current industry needs and failures.

**Video may be required**: this episode is focused on demonstrating uses of LLMs against various code. As such, listeners may want to watch the stream to see these uses rather than just listening. Also, Seth and Ken talk briefly at the beginning of the episode about a new tldr;sec project (thanks Clint!) called awesome secure defaults that lists out useful libraries and projects that are secure by default.

After a week of travel, Seth and Ken return to the podcast with a breakdown of their travel experiences at multiple conferences and teaching their first Practical Secure Code Review course using LLMs to enhance the methodology. This is followed by reinforcement of code review steps including library research, a discussion of the recent XZ backdoor, and an article reviewing LLM hallucinations when recommending libraries.

When Ken is away, the geeks will play. Seth is joined by podcast regular Stefan Edwards (@lojikil) to catch up on his recent work around threat hunting. This progresses into a discussion on threat intelligence and what is available for applications. A recent blog post on the utility of the CVE system spurs thoughts on the usefulness of published CVEs. Finally, opinions fly on authorization issues and how simple misconfigurations result in the many vulnerabilities or attack chains.

Ken and Seth are back to talk about the difference and competing priorities of Application and Enterprise Security. In short, recent news contends that Enterprise or Infrastructure security is lacking, whereas Application or Product Security is in a good state. This is followed by a discussion on supply chain security tools due to a recent analysis conducted by DoyenSec comparing false positives and negatives from the leading tools.

Ken and Seth return for another episode, starting out with pointers on getting into security and finding a niche, all based on a recently released Microsoft project to introduce anyone to security. This is followed by a discussion on Chinese hacking groups and recent breaches among those groups. Finally, a discussion protecting the software supply chain due to recent forking and upload of malicious repositories on GitHub.

Seth and Ken review the recent Whitehouse report on going back to the basics for software security and vulnerabilities. Specifically, how is the use of memory unsafe languages like C and C++ affecting the overall security of the internet landscape. This include a discussion on formal verification and crocs and socks of software testing. Finally, thoughts are shared on the recent use of Hugging Face and Github to host malicious code/packages and how this is a natural progression for popular package repositories.

Podcast viewers will be familiar with Portswigger's annual list of Web Hacking Techniques. Ken and Seth take some time to digest the list and recommend reviewing not only the top 10, but also the nominations. A discussion on the use of LLM Agents as a dynamic scanning engine for identifying vulnerabilities. If you aren't already using an LLM to help speed up your AppSec, why not? Finally, a discussion on security statistics and how bad they are.