Ep 23 - Part 2 - AI Security Incident Response, Supply Chain Chaos, AI Training and the Compliance Wake Up Call
Coffee, Chaos and ProdSecShow Notes
🎙️ Coffee, Chaos and ProdSec, Ep 23 AI security keeps getting talked about, but incident response, supply chain risk, and people are still treated like someone else’s problem. So this week, Kurt and Cameron grab their mugs and spend the episode walking through what actually happens when AI systems misbehave, agents start acting outside expectations, and traditional security playbooks stop lining up with reality. From AI specific incident response that doesn’t fit existing SOC workflows, to Software Supply Chain Security risks hiding in models, vendors, and embedded SaaS features, this episode gets into where things really break once AI is in production. They also dig into why AI training across teams matters more than most controls, how shadow AI keeps showing up in unexpected places, and why compliance pressure around AI is no longer theoretical as regulations and accountability get real. If you work in Cybersecurity, Application Security, Product Security, DevSecOps, or you’re trying to prepare your org for AI incidents, audits, and uncomfortable questions, this episode will sound very familiar. ☕ New episodes every Wednesday. Coffee, Chaos and ProdSec -> strong coffee, stronger opinions.
More from Coffee, Chaos and ProdSec
🎙️ Coffee, Chaos and ProdSec, Ep 60 Your build pipeline runs whatever gets pushed to it. Your secrets sit unmasked in runner memory. And the attackers figured this out before your defenders did. This week Cameron and Kurt bring on François Proulx, CTO of Boost Security, who spent four years publishing supply chain research only to watch threat actors quote his own tools back to him on a dark web forum. So he built SmokedMeat, Metasploit but for CI/CD, and a CTF called whooli so you can run it without going to jail. From the Ultralytics and tj-actions compromises, to dumping OIDC tokens straight out of an ephemeral runner, to why private repos are the bigger risk and not the safer one, this episode gets into how build pipeline exploitation actually works and what Product Security teams keep getting wrong. Real attack paths, real incidents, and the AI angle making all of it faster. If you work in ProdSec, DevSecOps, Security Architecture, or Cybersecurity and you still don't threat model your YAML, this one is for you. ☕ New episodes every Wednesday. Coffee, Chaos and ProdSec -> strong coffee, stronger opinions.
🎙️ Coffee, Chaos and ProdSec, Ep 59 AI writes most of the code now, and Cybersecurity still shows up after it ships to review it. Cameron and Kurt join Michael Nov, Co-Founder and CEO of Prime Security, to fight about where Product Security belongs in the agentic SDLC. Cameron wants scanning inside the agent loop and out of the PR. Michael says keep control at the PR, because a developer can just type ignore. They also get into whether developers should still learn security or whether you build guardrails around them, and who owns the fix when a vendor promised one and the breach landed anyway. Kurt's Security Architecture answer: the judge tells you to leave. If your Application Security scanner hands you thousands of findings and no fix, this DevSecOps chaos will feel familiar. ☕ New episodes every Wednesday. Coffee, Chaos and ProdSec -> strong coffee, stronger opinions.
🎙️ Coffee, Chaos and ProdSec, Ep 58 Your PR gate blocked the merge again. Or worse, the scan is still spinning and engineering wants to know why. This week Cameron and Kurt make the case for informing over blocking. Cameron explains why every PR blocking rollout he's run caused more headache than value, and it was the ops game, not the findings. Engineering merges, engineering owns the risk. Kurt takes it into the agentic SDLC, where scanners find and agents fix at machine speed, so a blocking gate just stalls the loop. They also vent about the chaos of the AI news cycle and call agent sandboxing what it is, system hardening. If you work in Product Security, Application Security, DevSecOps, Security Architecture, or anywhere in Cybersecurity and you've ever unblocked a PR at 2 AM, this one's for you. ☕ New episodes every Wednesday. Coffee, Chaos and ProdSec -> strong coffee, stronger opinions.
🎙️ Coffee, Chaos and ProdSec, Ep 57 Your AI chat window has a human reader now. Your bug bounty program can't triage its own findings. And somewhere out there, 18 AI agents are shipping code that nobody's going to review. This week Cameron and Kurt trace one thread through four different stories: the gap between what we assume is private and what actually is. From an encrypted AI chat pitch built on the same logic that created Signal, to a leaked contractor pipeline reading real ChatGPT conversations, to a vulnerability management rant about why finding bugs faster doesn't make you more secure, to a solo developer running a fleet of AI agents that write more code in a day than any human could review in a week. They cover zero retention agreements that nobody can actually audit, bug bounty programs that exist for legal cover more than for catching bugs, and what it means for Product Security and DevSecOps when code review quietly stops being a human job. If you work in ProdSec, Application Security, or Security Architecture and you've ever clicked "accept" without reading it, this one's for you. ☕ New episodes every Wednesday. Coffee, Chaos and ProdSec -> strong coffee, stronger opinions.
🎙️ Coffee, Chaos and ProdSec, Ep 56 Everyone agrees AI needs to slow down. Nobody is actually slowing down. Cameron's already annoyed before the coffee's poured. This week Cameron goes on a rant about people who let Claude build something and then post like they hand coded it themselves, and why he thinks agent skills are just long prompts dressed up to sound complicated. Kurt pushes back on where the line actually sits between prompting and ownership. Then they get into the AI slowdown discourse nobody can agree on, a governing agent that stood down because the model it was watching claimed to be "just in a simulation," and the NSA, CISA and FBI advisory naming six China based companies for distilling US frontier models, which Cameron calls corporate espionage through the API and Kurt says flat out smells funny. They close on GTIG's six hour agentic attack chain, why 30 day patch SLAs are already dead, micro segmentation making a comeback, and Kurt's callout that every cyber capability benchmark for these models gets run with the guardrails switched off. If you work in Cybersecurity, Application Security, Product Security, DevSecOps, or Security Architecture and you're sick of hearing "guardrails" used as a whole sentence, this one's for you. ☕ New episodes every Wednesday. Coffee, Chaos and ProdSec -> strong coffee, stronger opinions.