The Application Security Podcast

What happens when a protocol goes from idea to everywhere before security can catch up? Vandana Verma Sehgal, leader of the OWASP MCP Top 10, returns to explain why most organizations cannot say how many MCP servers are running on their developers' machines. She walks Chris and Robert through how MCP turns a chatbot into a new execution layer, how four lines of malicious code in a tiny server can slip past everyone, and why shadow MCP servers are shadow IT with direct access to action. The three debate whether we are missing the window to build AI security in from the start, why consent fatigue is undermining human in the loop, and where command execution really happens. Plus: why a secure MCP server should be boring, and why no MCP server deserves trust on day one.

This episode is sponsored by Security Compass. Make modern software development secure, consistent, and provable.

About Security Compass
AI writes code faster than anyone reviews the design. Threats do not wait for an annual assessment. Security Compass models threats continuously and turns them into requirements developers act on, not a report read after ship.
→ Learn more about securing the AI-DLC with Security Compass

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea
Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with Vandana Verma Sehgal:
→ Vandana Verma Sehgal on LinkedIn

Mentioned in this episode:
→ OWASP MCP Top 10
→ OWASP MCP Top 10 on GitHub
→ OWASP Top 10 for Large Language Model Applications
→ OWASP Top 10 for Agentic Applications for 2026
→ OWASP Agentic Skills Top 10
→ Model Context Protocol
→ EU AI Act (Regulation (EU) 2024/1689)
→ Singapore IMDA: Model AI Governance Framework for Agentic AI

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00:00 - Cold open: with great power comes great responsibility
00:00:20 - Meet Vandana Verma Sehgal
00:00:34 - Welcome and what gets Vandana away from screens
00:03:05 - What is MCP, and why should AppSec teams care now?
00:06:56 - MCP moved faster than any protocol before it
00:10:30 - Are we missing the window to secure AI?
00:12:31 - Harnesses, an AI security manifesto, and the EU AI Act
00:17:50 - Why the world needed an OWASP MCP Top 10
00:21:02 - Where command execution really happens
00:23:57 - How the list was built: community feedback and CVE data
00:26:46 - MCP09: shadow MCP servers
00:29:39 - If you could fix only one: scope creep
00:30:48 - Authentication, authorization, and delegated authority
00:35:24 - Consent fatigue and the human in the loop
00:37:04 - What a secure MCP server looks like: boring by design
00:41:27 - How AppSec teams should engage developers building MCP
00:44:46 - How to contribute to the OWASP MCP Top 10
00:45:41 - Key takeaway: never trust an MCP server on day one

What happens when a tsunami of AI-written code meets a security industry that has spent decades chasing vulnerabilities? Brook S.E. Schoenfield, CTO of Rezliant and author of Securing Systems, returns to warn about "the illusion of omniscient perfection," the cheerful confidence that makes AI output feel finished, and the "learned fragility" that sets in when nobody understands the code anymore. He, Chris, and Robert debate whether an AI threat model that gets 90% of the way there is a win, how Brook now uses AI for reachability analysis, and why prompt injection traces back to a 1975 design principle we ignored. They ask what architecture even means in the age of AI, where the next generation of architects will come from if nobody writes code, and why the collapsing cost of fixes means it's time to stop chasing vulnerabilities.

This episode is sponsored by Security Compass. Make modern software development secure, consistent, and provable.

About Security Compass
AI writes code faster than anyone reviews the design. Threats do not wait for an annual assessment. Security Compass models threats continuously and turns them into requirements developers act on, not a report read after ship.
→ Learn more about securing the AI-DLC with Security Compass

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea
Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with Brook S.E. Schoenfield:
→ Brook S.E. Schoenfield on LinkedIn
→ brookschoenfield.com

Mentioned in this episode:
→ Rezliant
→ Securing Systems: Applied Security Architecture and Threat Models by Brook S.E. Schoenfield
→ Saltzer and Schroeder: "The Protection of Information in Computer Systems" (1975)
→ Izar Tarandach on LinkedIn
→ Dwarkesh Patel: "The Rise and Fall of Agent Civilizations" (essay on the OpenAI–Hugging Face incident)
→ Kadrey v. Meta (2025 fair use ruling on AI training)

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00:00 - Cold open — The illusion of omniscient perfection
00:00:48 - Welcome back, Brook S.E. Schoenfield
00:01:51 - AI is writing code and filling the security skills gap
00:06:18 - Can AI threat model as well as the experts?
00:08:46 - Is 90% of the way there good enough?
00:13:47 - Closing one weakness beats chasing perfection
00:16:33 - When AI threat modeling is enough, and when it isn't
00:17:54 - Learned fragility
00:19:26 - Embrace securing AI or be left in the dust
00:23:41 - Against set and forget: agent drift and AI ethics
00:29:30 - What does architecture mean in the age of AI?
00:30:29 - Mixing control and data planes: the root of prompt injection
00:32:49 - Reachability: how Brook threat models with AI today
00:35:08 - Diagram drift and inclusive threat modeling
00:38:46 - Will developers disappear and architects emerge?
00:41:00 - The cost of code has collapsed
00:44:33 - Where will the next architects come from?
00:49:15 - Final thoughts: stop chasing vulnerabilities

Jim Manico thinks the era of human code review is ending, and that clinging to it will hurt your company. The founder of Manicode Security returns to explain why AI didn't kill AppSec education but supercharged it, why vague prompting on frontier models turns companies into "token furnaces," and why prompt injection is the one genuinely new vulnerability class of the AI era. He walks Chris and Robert through his full AI coding workflow, from reverse engineering an architecture file to security rules and planning-mode build plans that make AI output deterministic. The three debate whether AI will finally eliminate SQL injection, whether AppSec vendors can survive without integrating cyber models, and when humans still need to step in: the moment an agent tries something its policy doesn't allow. Plus: the one habit every security leader should teach developers now.

This episode is sponsored by Security Compass. Make modern software development secure, consistent, and provable.

About Security Compass
AI writes code faster than anyone reviews the design. Threats do not wait for an annual assessment. Security Compass models threats continuously and turns them into requirements developers act on, not a report read after ship.
→ Learn more about securing the AI-DLC with Security Compass

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea
Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with Jim Manico:
→ Jim Manico on LinkedIn

Mentioned in this episode:
→ Manicode Security
→ Manicode Forge
→ OWASP Artificial Intelligence Security Verification Standard (AISVS)
→ OWASP AISVS on GitHub
→ Claude Code
→ Ollama

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00:00 - Cold open: the era of code review is ending
00:00:47 - Meet Jim Manico
00:01:03 - Welcome and what gets Jim away from screens
00:05:29 - How AI changed developer security education
00:07:03 - Teaching developers to use AI well
00:09:26 - Where AI and AppSec stand: the token furnace
00:12:08 - Separating signal from hype
00:14:24 - Integrate with cyber models or die: the future of AppSec tools
00:18:34 - Are AI coding assistants creating new vulnerabilities?
00:20:02 - Prompt injection: the one truly new class
00:21:15 - Will AI eliminate the OWASP Top 10?
00:24:18 - Local models and Apple's hardware edge
00:28:32 - Jim's AI coding workflow, step by step
00:33:57 - The end of human code review
00:38:31 - Can we trust AI code review, and when do humans step in?
00:44:56 - Where companies really are with AI
00:47:02 - One change for security leaders: planning mode
00:48:19 - Wrap up

Three years ago, Jeevan Singh mapped out what an application security engineer needed to know. AI has rewritten the job since. Jeevan, Director of Security Engineering at Rippling, returns to unpack how his team polices thousands of engineers shipping 10x more code: a "vulnerability jail" that locks non-compliant teams out of the main branch, AI-reviewed extension requests, and homegrown agents that hunt for entire classes of vulnerabilities instead of one bug at a time. He and Chris debate whether AI has killed classic SAST and DAST, whether code review still needs a human in the loop, and whether bug bounty programs still make sense when the researchers on both sides are running the same models. Plus: one concrete move every AppSec leader can make this quarter.

This episode is sponsored by Security Compass. Make modern software development secure, consistent, and provable.

About Security Compass
AI writes code faster than anyone reviews the design. Threats do not wait for an annual assessment. Security Compass models threats continuously and turns them into requirements developers act on, not a report read after ship.
→ Learn more about securing the AI-DLC with Security Compass

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea
Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with Jeevan Singh:
→ Jeevan Singh on LinkedIn

Mentioned in this episode:
→ Rippling
→ Dwarkesh Patel: "The Rise and Fall of Agent Civilizations" (essay on the OpenAI–Hugging Face incident)

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00:00 - Cold open: vulnerability jail
00:00:55 - Meet Jeevan Singh
00:01:11 - Welcome and Robert's AI lab
00:02:37 - What gets Jeevan away from the machines
00:04:51 - Hardware projects with his son
00:06:20 - What's changed for the AppSec engineer since AI
00:08:17 - Shipping production code and fixing whole vulnerability classes
00:09:13 - Why AppSec has to become less collaborative
00:10:02 - From democratized vuln management to vulnerability jail
00:11:50 - How engineering reacted and the feature flag jail precedent
00:14:05 - From manual jail to automated checks
00:15:30 - An AI bot that reviews SLA extensions
00:16:06 - Can AI wipe out an entire vulnerability class?
00:17:36 - Building an anti-SSRF library and rolling it out
00:19:40 - Which AppSec skills matter now
00:22:28 - Validating all that AI-generated code
00:22:57 - Agents that hunt for vulnerability classes
00:24:38 - Is this the death of classic AppSec tools?
00:26:51 - Code review and humans in, on, and out of the loop
00:28:00 - Objective-based agents that find RCEs
00:28:59 - Build vs. buy for AppSec teams
00:31:29 - Advice for teams of one to five AppSec engineers
00:32:58 - Cutting SLAs to 3, 5, 7, and 10 days
00:34:31 - Parachuted in as the only AppSec engineer
00:37:59 - One investment to make this quarter
00:39:09 - The Hugging Face and OpenAI agent incident
00:40:13 - Is bug bounty dead?
00:42:19 - Key takeaways: be an engineer, run toward AI
00:43:53 - Wrap-up

Most fault trees get built on gut feeling. Petra Vukmirovic did something rarer: she borrowed the actual math from aviation and nuclear-plant safety engineering and pointed it at AI agents. Petra traded emergency medicine for application security and now heads information security at Numan — and she joins Chris Romeo and Robert Hurlbut to make the case for fault tree analysis (FTA), the deductive method that picks up exactly where threat modeling stops. Petra walks through a "wrong customer refund" AI agent scenario step by step, showing how AND/OR gates and minimal cut sets turn vague worry into ranked, data backed probabilities. They dig into where AI helps build a tree, and where garbage in, garbage out still applies, why "comprehensive test coverage" is a myth, and how attaching real dollar figures to failure paths makes it easier to sell security controls to leadership.

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea
Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with Petra Vukmirovic:
→ Petra Vukmirovic on LinkedIn
→ OWASP Threat Model Library

Mentioned in this episode:
→ Adam Shostack: "Stop Trying to 'Manage Risk'" (keynote)
→ OWASP Global AppSec USA 2026 (San Francisco, Nov 5–6)

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00 Cold open — the math behind where to put your controls
01:09 Meet Petra Vukmirovic
01:28 Petra's origin story: from ER doctor to AppSec
02:50 Career path: engineer to Head of InfoSec at Numan
04:18 What is fault tree analysis, and where threat modeling ends
06:22 Can AI actually do fault tree analysis?
08:12 Walking the "wrong customer refund" agent example
12:33 Storing your trees: JSON vs. Markdown
16:08 Why conjunctive failures trip up narrow thinking
17:27 Top 3 failure modes when agents touch downstream systems
19:29 Real story: an agent pushed code to main without approval
21:27 Testing: why "comprehensive coverage" is a myth
23:54 How rough is rough? Assigning probabilities
28:08 Getting started without a six week science project
31:35 Using FTA to sell controls and build credibility
33:43 The epiphany: FTA is about controls, not faults
34:48 The one thing every agentic team should add today
35:48 Closing thoughts and OWASP Global AppSec USA preview

Most security champions programs don't fail on tooling — they fail on people. Lisi Hocke spent three years as a champion before moving fully into product security, which means she has argued both sides of this from inside the trenches. Drawing on the talk she and Mireia Cano gave at OWASP Global AppSec EU 2026, Lisi walks us through the four things that actually make these programs work: psychological safety first, then cognitive load, then influence when you hold no formal authority, then a champions community so the whole thing doesn't stall the week security goes on vacation. We also get into cutting security wait times, winning organizational support, what AI does and doesn't change here, and why she will tell you never to record the champions meeting.

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea
Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with Lisi Hocke:
→ Lisi Hocke on LinkedIn
→ A Tester's Journey — Lisi's blog

Mentioned in this episode:
→ Slides: Security Champions — Lessons from Opposite Trenches (with Mireia Cano)
→ OWASP Juice Shop

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00 Cold open — what psychological safety actually means
00:56 Meet Lisi Hocke
02:25 Lisi's security origin story
05:42 "That place was taken" — becoming a champion anyway
07:39 Moving into a full-time product security role
08:39 Meeting Björn Kimminich, the Juice Shop project lead
09:23 Why role play instead of a normal conference talk
12:27 Security and development, disconnected
14:19 The first full-time security role
15:14 Making people wait is the real damage
17:10 Cutting the backlog and the turnaround time
19:31 What Lisi got dead wrong
20:08 What testing and quality work taught her
21:31 The four things that make champions programs work
22:07 One: fostering psychological safety
24:50 Champions without their manager's blessing
28:45 Two: managing cognitive load
29:46 Three kinds of load, and which one to cut
31:21 Three: power sources when you have no formal authority
33:03 Four: build a champions community
34:38 Keeping security people from burning out
36:37 How AI changes who you recruit and what you need
39:32 Should AI change champions programs at all?
40:33 Psychological safety when a bot joins the meeting
42:25 Don't record the champions meetings
43:26 Programs that outlive the person who started them
45:59 Key takeaway and homework
47:21 Closing thoughts

Is traditional DAST finally dead? James Berthoty came back to settle the argument that his last episode started. James is the founder and analyst behind Latio, and he argues that AI pentesting is a genuinely different animal — payloads generated with context about your actual application, agents that chase findings the way a human tester would, and results a scanner was never going to produce. We get into what it costs once tokens enter the picture, who pays for them, whether a pentest on every pull request is realistic, and what stops an autonomous tester from going further than it should. Then we look further out: the future of bug bounties, what happens when cloud and model providers absorb today's security tooling, and who is accountable when an agent deletes your production database.

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea
Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with James Berthoty:
→ James Berthoty on LinkedIn
→ Latio
→ Latio Pulse

Mentioned in this episode:
→ Latio's free reports
→ James on the podcast the first time: Is DAST Dead? And the future of API security

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00 Cold open — the results speak for themselves
01:01 Meet James Berthoty and the "Is DAST dead?" fallout
01:31 Chickens, eggs, and getting away from screens
03:46 Why we're revisiting the DAST question
04:14 A working definition of AI pentesting
05:47 Contextual payloads and application awareness
06:47 Determinism, repeatability, and what buyers actually want
07:46 Can you run an AI pentest on every code change?
09:43 What it really costs
10:40 Incumbents vs. AI-native vendors
13:27 Who pays for the tokens?
14:29 Bundling, platforms, and competitive pressure
16:25 AI across the whole development workflow
18:22 Agents that run all the way to deployment
19:21 A pentest on every pull request
21:52 What stops a pentest from going too far?
23:10 Permission scoping and guardrails
26:07 Where the findings actually land
28:02 The future of bug bounties
30:50 Why pentests command more budget than DAST
31:45 Could the cloud providers absorb security tooling?
34:38 What model providers could build instead
36:36 The same story on the code scanning side
39:24 Accountability when the tool misses something
40:22 Shared responsibility when an agent deletes production
41:23 The verdict on DAST
42:19 Where to find Latio's free reports
43:15 Closing thoughts

AI security has no shortage of standards — the problem is turning them into something a team can actually use. Rob van der Veer has spent 34 years in AI and security, founded the OWASP AI Exchange, and created MOSAIC, the agreement that brought eight standards bodies together with SANS to stop the fragmentation. Rob explains what responsible AI really means, what the EU AI Act actually asks of you, and why most AppSec teams are still missing the point on AI-generated code. We also get into agentic red teaming, what happens when agents quietly exceed their scope, and whether AI finally levels the playing field between attackers and defenders. If you build software with AI in it — or with AI — this one is worth your time.

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea

Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with Rob van der Veer:
→ Rob van der Veer on LinkedIn
→ OWASP AI Exchange
→ MOSAIC

Mentioned in this episode:
→ OpenCRE
→ Luna and the Magic AI Paintbrush

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00 Cold open — don't be surprised when the AI breaks out of the cage
01:15 Meet Rob van der Veer: music, cycling, and the Hoodoo 500
05:24 Defining responsible AI
07:41 Fairness, protected attributes, and transparency
09:34 The EU AI Act and what regulation actually asks of you
11:27 How AI changes every part of software development
13:23 Where responsibility lands
15:20 You're not defending your own data center
17:19 What traditional AppSec teams consistently miss about AI
18:18 Finding vulnerabilities in AI-generated code
19:19 Too many standards — and using AI to write them
20:51 MOSAIC: eight standards bodies, one agreement
22:09 One machine-readable taxonomy with OpenCRE
23:06 Can AI level the field between attackers and defenders?
25:59 When AI security becomes security theater
26:56 What agentic red teaming actually looks like
29:44 When agents exceed their scope
32:43 Luna and the Magic AI Paintbrush
33:40 Do we sandbox the agents?
34:40 Guardrails without killing creativity
36:39 Skill atrophy when AI is your only way forward
40:04 "How do we hit this quarter?" and the pressure to ship
43:16 Everyone is selling agentic security
45:07 Key takeaways and where to start with the AI Exchange
47:12 Closing thoughts

You don't have to let AI do the thinking for you. In this episode, Vikram Narayan shares why the smartest teams use AI as an accelerant — not a replacement — and why human judgment still matters most in threat modeling. Vikram created Precogly, an open-source threat modeling platform now running as an OWASP project, and he walks us through what it took to build a free tool on par with commercial vendors. We dig into the tension among speed, compliance, and real risk; whether the Threat Modeling Manifesto needs amending for AI; and what it means to "fight the AI" so critical thinking stays sharp. If you care about AppSec, AI, and the future of threat modeling, this conversation will give you a lot to think about.

This episode is sponsored by Corgea. Design it. Build it. Ship it. Corgea secures it.

About Corgea
Corgea is an AI-native application security platform that secures software from design to production. It brings together security design reviews, AI SAST, dependency and IaC scanning, code quality checks, and autonomous pentesting—helping security and engineering teams find risk earlier, fix what matters, and ship securely.
→ Learn more about Corgea

Connect with Vikram Narayan:
→ Vikram Narayan on LinkedIn
→ Precogly — open-source threat modeling (OWASP project)

Mentioned in this episode:
→ Threat Modeling Manifesto
→ ThreatModCon

Follow the Application Security Podcast:
➜ Home: appsecpodcast.com
➜ X: @AppSecPodcast
➜ LinkedIn: The Application Security Podcast
➜ YouTube: @ApplicationSecurityPodcast
➜ Instagram: @appsecpodcast
➜ Facebook: Application Security Podcast

Chapters:
00:00 Cold open — the threat model that "feels wrong"
01:22 Welcome and introductions
02:17 Vikram's security origin story
05:43 From machine learning research into LLMs
06:42 Hospital chatbots, hallucination, and knowing when to escalate
07:51 ThreatModCon and the case for an open-source threat modeling tool
09:35 IoT, emergence, and the traffic-light problem
11:17 The OWASP Vienna talk and the Threat Modeling Manifesto
12:26 Why "AI, just do the threat model" falls apart
15:14 What AI is actually good at in threat modeling
18:07 Human discomfort vs. the machine's confident answer
20:29 Inside Precogly: accelerant, not replacement
20:58 Library packs and the skills layer
24:50 Where AI kicks in — and where it shouldn't
27:48 Should the Threat Modeling Manifesto be amended for AI?
30:28 Where Chris and Robert land
31:36 Wi-Fi sensing, privacy, and modeling what you can't see
33:15 If you can't explain it, can you trust it?
35:11 Beyond checklists — design-level questions
35:59 Fight the AI — Vikram's key takeaway
39:10 Closing thoughts

AI is moving AppSec's control point out of CI and directly into the coding agent—but what happens when the model writing the code is also expected to secure it? Semgrep co-founder and CEO Isaac Evans explains why deep background analysis and real-time agent plugins may replace universal rule sets with organization-specific security controls. He and Chris explore how security engineering roles will change, why independent verification still matters, and where business-logic flaws may become the next major battleground. The conversation also covers vibe coding at enterprise scale, the limits of reasoning about model behavior, open source in an agent-built world, and why Isaac sees more opportunity than threat even as AI creates a fresh wave of vulnerabilities and cleanup work.

Connect with Isaac Evans:
→ Isaac Evans on LinkedIn
→ Semgrep

Mentioned in this episode:
→ Semgrep
→ DeepSeek
→ Cursor
→ OpenAI Codex
→ Claude Code
→ Boston Dynamics
→ DARPA Robotics Challenge
→ Reflections on Trusting Trust
→ uutils/coreutils
→ Rust

Follow the Application Security Podcast:
➜ Home
➜ X
➜ LinkedIn
➜ YouTube
➜ Instagram
➜ Facebook

Chapters:
00:00 Meet Isaac Evans
01:11 From cryptography to the DARPA Robotics Challenge
03:43 Founding Semgrep
04:05 How AI is reshaping AppSec
06:15 Attackers, defenders, and model choice
08:02 Regenerating code until it clears the security bar
10:30 Organization-specific rules beat universal rules
14:18 The changing role of the security engineer
17:53 Career advice for security practitioners
19:47 Will foundation models absorb security vendors?
24:18 Getting secure changes across an enterprise
26:40 Trusting Trust becomes the easy problem
27:41 How much should we trust agent-generated code?
29:38 Independent verification and competing models
34:50 Business logic flaws after SQL injection
36:56 Protecting the new wave of citizen developers
39:40 Vibe coding and disposable software
42:05 Open source in an agent-built world
44:10 Can the exponential pace continue?
44:41 Key takeaways and calls to action