On Episode 160 of the Silver Bullet Security Podcast, BIML’s Gary McGraw hosts Aaron Bedra. Aaron talks about enterprise adoption of AI/ML from his perspective at the intersection of security engineering and executive leadership; the shocking pace of ML evolution; ML grinding on the security exploit front; MLsec (including recursive pollution); and the problem of Agentic swarm control.  MLsec meets the real world.

 * Aaron Bedra - DRW | LinkedIn [https://www.linkedin.com/in/aaronbedra]
 * Programming Clojure, Third Edition [https://pragprog.com/titles/shcloj3/programming-clojure-third-edition/]
 * abedra (Aaron Bedra) · GitHub [https://github.com/abedra]
 * BIML on recursive pullution [https://berryvilleiml.com/2026/01/10/recursive-pollution-and-model-collapse-are-not-the-same/]
 * Silver Bullet episode 097: Aaron Bedra [https://apothecaryshed.files.wordpress.com/2018/10/silverbullet-097-aaron-bedra.mp3], 4.30.14

160 - Aaron Bedra

On Episode 159 of the Silver Bullet Security Podcast, BIML’s Gary McGraw hosts Melanie Mitchell [https://melaniemitchell.me/].  Melanie talks about the surprising progress made in AI since the '90s, whether real concepts emerge from data scale alone, whether modern connectionist systems extrapolate or only interpolate, and what humans do differently.  We also talk about the complex innards of modern transformer networks, measuring alien intelligences, role playing, sycophancy, and security, and why giving answers is easier than posing interesting questions.

 * FARG and Douglas Hofstadter [https://en.wikipedia.org/wiki/Fluid_Concepts_and_Creative_Analogies]
 * Copycat [https://mitpress.mit.edu/9780262132893/analogy-making-as-perception/]
 * The NeuriPS keynote [https://neurips.cc/virtual/2024/109414]
 * BIML's measurement work [https://berryvilleiml.com/docs/no-security-meter-ai.pdf]
 * Santa Fe Institute [https://www.santafe.edu/]

159 - Melanie Mitchell

On Episode 158 of the Silver Bullet Security Podcast, BIML’s Gary McGraw hosts Artem Dinaburg [https://www.dinaburg.org/about.html].  Artem talks about using Agentic AI to find, fix, and exploit software security defects.  We talk decompilation, stochasticism and the tension between harness development and LLM nondeterminism, human intuition, and the hard parts of program analysis.  We also talk about MLsec (machine learning security), recursive pollution, and tokenization.  Do we need more people in software security in the age of AI?  Yes we do.

 * Trail of Bits [https://blog.trailofbits.com/]
 * Mastadon [https://mastodon.social/@trailofbits@infosec.exchange]
 * Evaluating Solidity support in AI Coding Assistants [https://blog.trailofbits.com/2024/11/19/evaluating-solidity-support-in-ai-coding-assistants/]
 * Researchgate [https://www.researchgate.net/scientific-contributions/Artem-Dinaburg-2148442770]

158 - Artem Dinaburg

On Episode 157 of the Silver Bullet Security Podcast, BIML’s Gary McGraw hosts Tim Schulz [https://www.linkedin.com/in/tim-schulz].  Tim talks about whitebox control and observability in machine learning systems (and especially transformer architectures), the limits of red teaming for securing AI,  "neural surgery,"  Agentic AI and the confused deputy problem, and the economics of network "smallification."

 * Starseer [https://starseer.ai/home]
 * Whitebox machine learning and looking inside networks [https://starseer.ai/blog/opening-the-hood-detecting-architectural-backdoors-in-ai-models]
 * Anthropics circuits thread [https://transformer-circuits.pub/]
 * Agentic AI and intention [https://berryvilleiml.com/results/no-security-meter-ai.pdf]

157 - Tim Schulz

On Episode 156 of the Silver Bullet Security Podcast, BIML’s Gary McGraw hosts Phil Venables [https://www.philvenables.com/about].  Phil talks about the evolution of the CISO role from running an engineering team in the '90s back into running an engineering team in the mid-2020s, Agentic AI and tools using tools, the rise of machine learning security, when we might see an AI BSIMM, emergent computation and security control, and what role humans can play in AI rollout.

 * Phil Venables at Ballistic [https://ballisticventures.com/team/phil-venables/]
 * Scaling Security, AI, and more [https://%20https//www.philvenables.com/post/scaling-security-ai-and-more]
 * Cloud CISO Perspectives [https://cloud.google.com/blog/products/identity-security/cloud-ciso-perspectives-phil-venables-on-ciso-2-0-and-the-ciso-factory]
 * BSIMM archive [https://www.garymcgraw.com/technology/bsimm/], coalition for secure AI [https://www.coalitionforsecureai.org/]

156 - Phil Venables

On Episode 155 of the Silver Bullet Security Podcast, BIML's Gary McGraw hosts Giovanni Vigna [https://sites.cs.ucsb.edu/~vigna/].  Giovanni talks about the evolution of software exploit and AI, what has happened in the security field since 2011 (and episode 65 [https://apothecaryshed.com/wp-content/uploads/2018/10/silverbullet-065-giovanni-vigna.mp3]), getting inside connectionist networks to see what is actually happening, and what it means to be a human hacker with intuition and intention.

 * Giovani Vigna [https://sites.cs.ucsb.edu/~vigna/]
 * iCTF [https://ictf.cs.ucsb.edu/]
 * Artiphishell [https://artiphishell.com]
 * The ACTION institute [https://action.ucsb.edu/]

155 - Giovanni Vigna

On Episode 154 of the Silver Bullet Security Podcast, Gary McGraw hosts Gadi Evron.  Gadi talks about the [un]prompted conference and its focus on fundamental aspects of machine learning security. The conversation touches on current approaches to MLsec, the words we use to describe AI security issues, and how quickly the field is evolving.

Links

 * Gadi Evron https://www.knostic.ai/about-us
 * [un]prompted https://unpromptedcon.org/
 * badness-ometer https://blog.citp.princeton.edu/2006/02/16/software-security-badness-ometer/
 * recursive pollution https://berryvilleiml.com/2026/01/10/recursive-pollution-and-model-collapse-are-not-the-same/

154 - Gadi Evron

Welcome back to the Silver Bullet Security Podcast, created and hosted by the Berryville Institute of Machine Learning [https://berryvilleiml.com/], where we are defining the future of machine learning security.  From 2006-2018, we explored the nascent field of software security through the lens of building security in.  But today, the frontier has moved.  As we integrate machine learning into the fabric of our essential systems, we find ourselves facing a new set of architectural flaws and security challenges that traditional software security can’t touch.  On Silver Bullet, we’re shifting our focus to the security of machine learning—bringing the same deep-dive, “no silver bullet” philosophy to the world of AI.

The first (154th) episode launches soon.

Silver Bullet Security Podcast promo

Silver  Bullet Security Podcast (@rmacmich) · Zencastr

Silver  Bullet Security Podcast (@rmacmich)

Welcome to the Silver Bullet Security Podcast, created and hosted by the Berryville Institute of Machine Learning [https://berryvilleiml.com/], where we are defining the future of machine learning security.  From 2006-2018, we explored the nascent field of software security through the lens of building security in.  But today, the frontier has moved.  As we integrate machine learning into the fabric of our essential systems, we find ourselves facing a new set of architectural flaws and security challenges that traditional software security can’t touch.  On Silver Bullet, we’re shifting our focus to the security of machine learning—bringing the same deep-dive, “no silver bullet” philosophy to the world of AI.

https://www.garymcgraw.com/technology/silver-bullet-podcast/

Theme song is "Zagreb" by The Cheebacabra (https://linktr.ee/cheebacabra).