
Jade sits down with not one but three Oak Security auditors, Bernd, Christian, and Mario, to debrief on DevCon in Mumbai and talk honestly about where auditing, AI, and security are heading.
The DevCon read is a little contrarian. The crowd was smaller and quieter than Bangkok, with fewer familiar faces and a lot of new Indian students, but the people who came were serious builders, and the bear market gets some credit for clearing out the noise. The irony of the week: everyone on stage was talking post-quantum and formal verification, while the Binance hack, which started with a laptop and a fake job offer, was unfolding in real time. As Christian puts it, there’s a gap between the shiny things we discuss and the boring ways people actually get rekt.
That gap is the backbone of the episode: the industry got good at securing smart contracts, and attackers simply moved to the operational layer. Compromised laptops, fake job offers, the “spy girlfriend,” social engineering. The panel is candid that new teams, who never lived through Web2 security, treat code that works as code that’s safe, and that AI has widened the attacker-defender asymmetry, a black hat can now monitor your GitHub, have an LLM review the diff of a security fix for a few tokens, and exploit it before a decentralised protocol can even ship the patch. Honestly, though, they admit the demand for operational-security audits still isn’t there; clients overwhelmingly want on-chain reviews, and we’re all feeling the results.
On AI in their own work, the three broadly agree it’s a genuine productivity multiplier, Bernd uses it to get up to speed on a codebase fast, Christian loves that POCs that took days now take minutes, Mario frames it as “just a tool, like a chainsaw,” powerful but dangerous without skill (“AI can find any bug, you just have to tell it exactly where it is”). They get into the model-versus-harness debate (Mario bets on the stronger raw model; Christian defends harnesses for orchestrating multiple models to argue findings out), the bimodal quality of AI-written code (either slop full of criticals or driven to perfection by skilled devs, with “human in the loop” as the real secret), and the quiet dread of AI taking the enjoyable parts of the job and leaving only the hard thinking. There’s also a genuinely useful tangent on keeping a cool head, the attacker’s edge is speed, so the defender’s edge is a checklist that stops you panicking.
The panel closes on what every team should change for 2027: take real responsibility and skin in the game (Bernd), grow up from anonymous dev groups into organised companies with real roles (Christian), and finally take operational security seriously, secure your infrastructure, protect your keys, understand your dependencies, and stop assuming “decentralised and audited” is enough (Mario). Or, as Bernd puts it: just assume you’re already compromised.
Key Topics
The DevCon debrief: the post-quantum conversation vs the Binance hack reality
Why operational security is the new frontier, and why clients still aren’t asking for it
AI in auditing: productivity, the model-vs-harness debate, and the future of the job
Chapters
00:00 Introduction and Meet the Three Auditors
02:14 The DevCon Debrief: Mood, Mumbai, and Fewer Tourists
03:56 The Binance Hack During a Post-Quantum Conference
06:03 Are Teams Asking About OPSEC, or Still “Our Code Is Fine”?
08:05 One Idea That Stuck: AI and the Attacker-Defender Asymmetry
09:57 Watching the GitHub: Exploiting Fixes Before They Ship
11:29 Mature Teams vs Newcomers, and False Confidence
15:07 Is Anyone Actually Asking for Operational-Security Audits?
17:52 Ground-Level OPSEC Tips: Bad UX, Checklists, and Updates
20:29 The DevCon Narrative: Privacy, Post-Quantum, and Minimising Trust
22:02 What “Minimising Trust” Means for the Key Holder
24:35 AI in the Audit: Assistant, Colleague, or Tool?
29:45 The Quality of AI-Written Code: Slop or Perfection
33:21 Chasing the Next Model, and When to Ship
36:17 Model vs Harness, and the Ideal Human-AI Division
43:11 The Future Role of the Auditor
47:16 The Future of Work Itself
48:21 One Thing Every Team Should Change for 2027
51:00 Where to Find Bernd, Christian, and Mario
Resources and Links
Christian Vari on X / LinkedIn: Christian Vari https://x.com/christianvari_
Mario Poneder on X / LinkedIn / GitHub: Mario Poneder https://x.com/marioponeder
Bernd on X: https://x.com/bernd_eth
Oak Security’s Research: https://research.oaksecurity.io/
Otros episodios de "CypherTalk"



No te pierdas ningún episodio de “CypherTalk”. Síguelo en la aplicación gratuita de GetPodcast.








