Open Models Will Do Anything: AI Is Speeding Up Attacks | Rob Chapman, ArmorCode

Joe's site: https://south-security.comNewsletter: https://south-security.com/newsletterRob Chapman of ArmorCode joins Joe South on Security Unfiltered.Rob took the long way into security, from bulletin boards in middle Tennessee to running security for oil and gas companies. Now he says anyone claiming AI isn't speeding up attacks isn't paying attention. Stock open models like DeepSeek push back once, then say okay, sounds good. They get into exploits weaponized within 30 minutes, why most exploitation comes through features and bad config instead of CVEs, why "patch everything in two weeks" isn't a program, and why the only way to keep up with vulnerability response is your own army of bots. Plus their personal AI agent setups, getting the data layer right, and why a year ago nobody wanted AI touching remediation and now everyone does.ArmorCode: https://www.armorcode.com/Rob Chapman: https://linkedin.com/in/rschapmanRob's site: https://rschapman.com/Show site: https://www.securityunfiltered.comAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

Joe's site: https://south-security.comNewsletter: https://south-security.com/newsletterShimon Tolts, co-founder and CEO of Copperhelm, AWS Hero, and CNCF and Claude Ambassador, joins Joe South on Security Unfiltered.Shimon argues 2026 collapsed the remediation window: AI writes exploits on demand, so "is there a public exploit?" is the wrong question and severity-based prioritization is dead. They get into FedRAMP shrinking critical remediation from 30 days to 12 hours, validating reachability and runtime instead of trusting CVSS, tailored WAF and eBPF policies as compensating controls, and why autonomous remediation is a people and process problem more than a tech one. Plus whether we're the last generation of coders, why GitHub and Git are hitting their limits, AI loops that improve themselves, token costs on the API side, and how AI is changing who gets to build.Shimon Tolts: https://www.linkedin.com/in/shimontolts/ (verify)Copperhelm: https://copperhelm.com (verify)Show site: https://www.securityunfiltered.comAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

Joe's site: https://south-security.comNewsletter: https://south-security.com/newsletterBill Gurling and Cal Al-Dhubaib of Rubrik join Joe South on Security Unfiltered to break down the OpenAI and Hugging Face incident: agents given impossible tasks that found each other through Artifactory, called themselves "the swarm," chained zero days up to cluster admin and broke into Hugging Face.From there they get into why agents act like relentless optimizers, not villains: the coding agent that exploited a Postgres CVE instead of asking for a password, the "make me coffee" problem, and how agentic search can leak sensitive data through a simple web query. Bill and Cal explain intent-based controls, why you need deterministic and LLM-based guardrails together, why your approved agents are a bigger risk than shadow AI, and why identity and authorization for agents has to become short-lived and scoped per session. Plus Cal's hottest take: we're going to need a lot more humans.Rubrik: https://www.rubrik.comShow site: https://www.securityunfiltered.comAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

Matt Lea, founder of Schematical and an AWS consultant since 2010, joins Joe South on Security Unfiltered.From dropping out of college to crash a startup, to calling a CEO straight from his WHOIS record, Matt explains why he bet his career on infrastructure, and why vibe coding is creating a whole new class of breaches. They get into the OpenAI agent that escaped its sandbox, model distillation, stolen API keys, an AI agent that posted environment data to Postbin, and the Terraform change that wiped a production database and its backups. Plus AWS vs Azure vs GCP, why Azure MFA took 12 checks to verify, Cloud War Games for training incident responders under pressure, the "lone wolf programmer" problem, and where to focus if you're breaking into cloud and AI security now.Chapters00:00 Seven months in the making (and a tornado)01:40 Matt's path: Q Basic, a failed startup, AWS in 201004:03 Why infrastructure is the steady career04:51 AI security is cloud security (for now)05:44 Vibe coding and the mid-market security gap06:39 "AI will take your job" vs reality08:22 Organized crime goes cyber12:28 Startups, scale and constant attacks13:24 OpenAI's agent escaped the sandbox18:02 Models are now worth stealing19:09 Model distillation explained23:35 Stolen API keys and vibe coding27:00 AI agents gone wrong: Postbin and a deleted prod database30:17 Why AWS (and when GCP)32:03 Azure frustration vs AWS34:37 GCP security by default37:46 Twelve places to check MFA in Azure39:35 Security spend vs runway40:47 Cloud War Games: training under pressure44:01 Tabletops, runbooks and the lone wolf programmer48:47 Where to start learning cloud and AI security51:50 Where to find MattMatt Lea / Schematical: https://schematical.comLone Wolf Programmer comics: https://schematical.com/comicsTech Debt (Steam, playtesters wanted): search "Tech Debt" on Steam (verify link)Subscribe: https://www.youtube.com/@securityunfilteredpodcastShow site: https://www.securityunfiltered.comAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

Joe Grand, aka Kingpin, hardware hacker, L0pht alum, and the engineer behind the $2M Trezor wallet recovery, joins Joe South on Security Unfiltered.They get into how a kid with an Atari 400 ended up in the hacker underground, why the hacker mindset can't be taught, and what 23 years as an independent hacker really looks like. Joe Grand walks through learning fault injection to free a locked wallet, how he and his colleague found an attack on STM32 chips that happens before the boot ROM even runs, why a vendor that won't respin silicon leaves chips vulnerable forever, and why "they need physical access" is not a security strategy (see: the San Francisco parking meter he bought on eBay). Plus failure, burnout, full disclosure, and his new travel show exploring hacker culture around the world.Chapters00:00 Finally reaching Joe Grand (email still wins)03:29 Why Security Unfiltered exists05:36 Atari 400 to the hacker underground10:00 The hacker mindset you can't teach13:16 Going independent for 23 years16:30 Keep learning or get cut21:04 Failing thousands of times with SELinux24:16 Hacks under pressure: cameras, customers, $66M27:55 How the $2M Trezor wallet hack started30:46 Learning fault injection34:37 Boot ROMs, Ghidra and silicon you can't patch37:04 Full disclosure and fighting for the user43:53 Chips that stay vulnerable forever44:34 Physical access is not security: the parking meter hack48:02 How an independent hardware hacker finds work52:28 Saying yes, saying no, and publishing your work56:22 Joe Grand's new hacker culture travel show62:57 Where to find Joe GrandJoe Grand on YouTube: https://www.youtube.com/@JoeGrandJoe Grand: https://joegrand.com/Grand Idea Studio: https://grandideastudio.com/Subscribe: https://www.youtube.com/@securityunfilteredpodcastShow site: https://www.securityunfiltered.comAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

Ashish Rajan is back on Security Unfiltered, a few years after we last recorded together at RSA. Ashish is a CISO, host of the Cloud Security Podcast and AI Security Podcast, and author of the new Wiley book AI Security Engineering.We get into how you write a book on a field that changes every month, why agents made him rewrite his plan halfway through, and what actually matters when you have millions of agents in your environment. Ashish makes the case that logs alone can't tell you intent ("is Joe an employee?" and "can I have Joe's password?" both look like plain text), why behavior is where agent security is heading, and why no vendor is going to build the internal context layer your security team needs. We also talk open weight models vs paying for tokens, whether the government had this tech first, and why human life still comes before technology.Chapters00:00 Ashish is back on Security Unfiltered02:40 Joe's dissertation: zero trust for satellites and post-quantum crypto06:36 Ashish's new book, AI Security Engineering07:07 Ashish's path: IAM, cloud security, Cloud Security Podcast08:26 From media company to AI advisory and the AI Security Lab10:19 Local compute, open weight models, Grok vs Claude13:35 Build harnesses you can swap out14:34 How do you write a book on a topic that changes every month?19:05 When research gets published faster than you can cite it20:54 Why network security is back in the conversation21:23 Agents breaking out of a sandbox22:57 Is the government already ahead on AI?23:54 The Manhattan Project comparison25:13 How cyber warfare teams test exploits26:41 Security for AI is a new field28:16 Human life over technology: Waymo and surgical robots29:58 Which agent security capabilities actually work?31:40 Make the agent explain why it pulled 10 records32:50 Behavior and intent over logs35:12 The internal context layer no vendor will build38:47 Logging is legacy at AI speed40:20 Horses vs cars: who gets left behind with AI42:17 Wrap upAshish Rajan on LinkedIn: https://www.linkedin.com/in/ashishrajan/AI Security Engineering (Wiley): https://www.amazon.com/dp/1394387687TechRiot: https://techriot.ioCloud Security Podcast: https://www.cloudsecuritypodcast.tvSubscribe: https://www.youtube.com/@securityunfilteredpodcastShow site: https://www.securityunfiltered.comX: @SecUnfPodcastAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

SPONSORED EPISODE: This episode includes paid promotion. Security Unfiltered was paid by Eve Security for this sponsorship.Joe sits down again with Nadav Cornberg (Eve Security) for the final episode in this Eve arc. They dig into agent runtime security: understanding intent inside a session, session tainting when privileges should shrink or grow, MCP connections that let tools take action in systems the author never planned for, and why static guardrails cannot keep up once agents start chaining actions on their own.Also covered: auto-remediation in AI studios, global policy packs, and Eve's AI prevention path that pushes agentic policies into tools you already run.Part 1: https://www.youtube.com/watch?v=3oklYF1iVLsEve Security: https://eve.security/Nadav on LinkedIn: https://www.linkedin.com/in/nadav-cornberg/Free AI risk assessment via Eve's contact form on eve.security.Subscribe: https://www.youtube.com/@securityunfilteredpodcastShow site: https://www.securityunfiltered.comX: @SecUnfPodcastAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

Antivirus is built to protect the device, not to judge a file. OPSWAT founder and CEO Benny Czarny explains why file scan detection lands somewhere between 5% and 40% even with AI engines, why 99.9% detection isn't good enough for a nuclear plant, and how Content Disarm and Reconstruction (CDR) protects critical infrastructure by assuming every file is malicious and rebuilding it clean.We get into:- How Benny bootstrapped OPSWAT and built a common language that let antivirus, VPN, firewall, and encryption products talk to each other- Testing every antivirus on the market with live malware, and what it showed about real-time protection vs file scanning- Why multiscanning with 30 engines still let malware through- How CDR rebuilds JPEG, PDF, and Office files to strip buffer overflow exploits before they land- Data diodes, one-way optical firewalls, and securing file flow into air-gapped networks- Offline patching, and why critical infrastructure teams need their own training (OPSWAT Academy)Benny's book, Cybersecurity Upside Down, covers the statistics and architecture behind all of this.Find Benny and OPSWAT:OPSWAT: https://www.opswat.comBenny Czarny: https://www.bennyczarny.comOPSWAT Academy: https://learn.opswatacademy.comChapters:00:00 Welcome Benny Czarny02:22 How Benny got into cybersecurity04:58 Big company vs small company lessons08:20 First hires and scaling from generalists to experts08:59 The problem that started OPSWAT12:01 Testing every antivirus with live malware13:42 Why antivirus file scanning fails, even with AI17:36 Multiscanning file flow into critical infrastructure19:45 Why 99.9% detection isn't good enough20:09 Assume every file is infected: CDR explained23:52 How CDR strips buffer overflow exploits26:50 Dolly the sheep and distilled water analogies28:16 From 15 file types to hundreds of formats29:51 Data diodes and one-way optical firewalls32:05 What CDR can't rebuild: executables and sandboxing33:48 Walled gardens and unpatched networks35:32 Offline patching and OPSWAT Academy39:12 Where to find Benny and his book#cybersecurity #criticalinfrastructure #OTsecurityAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

Joe and Danielle discuss how AI has accelerated an already messy SaaS security problem, especially as employees connect more tools, more agents, and more data without centralized oversight. The conversation focuses on why traditional block or allow controls are breaking down, and how security teams can use agentic automation to reduce risk without slowing the business00:00 - Why AI adoption is moving faster than cloud and SaaS ever did01:20 - The professional pressure to skill up on AI now03:19 - How cloud security and SaaS security lagged behind adoption06:13 - Why organizations can’t simply say no to AI tools08:30 - The difference between cloud-era experimentation and today’s browser-based AI10:37 - Why AI security and SaaS security are becoming the same problem12:31 - Why traditional onboarding and vendor review processes don’t scale14:43 - The Salesforce example that exposed hidden risk in a large enterprise17:00 - Why most third-party risk programs only cover part of the estate19:07 - How decentralized tech adoption bypasses security workflows20:31 - Why OAuth grants and third-party integrations are a major attack path22:25 - Why attackers usually go after credentials and tokens, not zero days23:19 - How AI lowers the barrier to entry for attackers26:55 - Why defenders need agentic capabilities too28:16 - The scale of unmanaged OAuth grants inside enterprises30:26 - Why no one can hire enough people to review every grant manually31:53 - How agents can analyze risk and recommend revocation at scale33:33 - The reality check from customer conversations about AI visibility35:24 - How security people think when told they can’t do something38:16 - Reactance theory and why employees work around controls40:25 - Incentives matter more than policy slogans41:48 - Why binary block or allow controls create shadow IT43:30 - How Nudge Security approaches SaaS and AI as one workforce problem46:14 - Why a workforce edge model matters more than network or identity alone48:26 - What just-in-time policy decisions could look like in the browser50:10 - Why policy experience is as important as policy enforcement52:08 - Danielle on Nudge Security’s free trial and LinkedIn presenceAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast

Joe talks with Dominik Richter, co-founder of Mondu, about how AI is changing security work, vulnerability management, and the expectations placed on technical teams. The conversation digs into why so many security workflows still produce noise instead of action, and why speed without quality creates more problems than it solves. They also explore what this means for hiring, junior talent, and the skills that will matter most as AI becomes embedded in day-to-day work.00:00 The Evolution of AI and Its Impact06:16 Challenges in Vulnerability Management13:00 The Role of AI in Security20:19 The Future of Work in an AI-Driven World21:09 Enhancing Productivity with AI29:41 The Evolving Role of Security Professionals40:16 Navigating the AI Landscape in Security48:52 Adapting to a Rapidly Changing Tech EnvironmentAffiliates➡️ OffGrid Faraday Bags: https://offgrid.co/?ref=gabzvajh➡️ OffGrid Coupon Code: JOE➡️ Unplugged Phone: https://unplugged.com/Unplugged's UP Phone - The performance you expect, with the privacy you deserve. Meet the alternative. Use Code UNFILTERED at checkout*See terms and conditions at affiliated webpages. Offers are subject to change. These are affiliated/paid promotions.Tesla Referral Code: https://ts.la/joseph675128Follow the Podcast on Social Media!Instagram: https://www.instagram.com/secunfpodcast/Twitter: https://twitter.com/SecUnfPodcast