DEF CON 34 - Main - Defender’s Dilemma Releasing Dual-Use AI Models as Capabilities Climb - Panel
Source: YouTube · DEFCONConference · published Nov 6, 2026 · 1:13:48
This Defcon panel with Anthropic's Jason Clinton, Bruce Schneier, and Google's Heather Atkins examines the "defender dilemma" of dual-use AI: you can't have AI cyber defenders without enabling attackers, and defenders face a turbulent 18–24 months ahead 4:10, 13:41.
Key Takeaways:
• Schneier frames dual use as universal and argues AI-powered defense is the only lasting friction, since scarcity-based controls failed for encryption and open/local models shed guardrails 4:10, 11:37.
• Clinton predicts an open-weights model with Mythos-class offensive capability within 3–6 months; Atkins proposes friction via availability, capabilities, legal consequences, and defender neutralization 13:41, 9:47.
• Alignment is underrated: Anthropic disclosed a model that halted an attack on its own; benchmark alignment like exploits 30:00.
• Recursive self-improvement is underway—8x code output per person, over half of Anthropic's code agent-written 36:16.
• Calls to action: urgency on the coming phase change (Clinton), integrity as the decade's defining problem—"integris" (Schneier), and sharing prototypes (Atkins) 1:09:05.
Consensus: get through the next two years with basics and AI-assisted defense, and we approach zero-day-free software.
Sources:
Generate CPE Credits
Generate a professional CPE document from this video's transcript.
Estimated credit: 1 CPE hours
Estimate uses the video runtime (1 hour ≈ 1 CPE, rounded to the nearest 0.5, minimum 0.5, maximum 2.0). The final amount can be lower after review, never higher.
Topic: Cybersecurity. Commonly maps to: Security and Risk Management, Security Operations. Exact CISSP domains are assigned during generation.
Free account. One generation at a time, with a daily limit.
CPEBuddy is independent and not affiliated with or endorsed by ISC2, ISACA, or any certification body. Exports are formatted for common CPE submissions; acceptance is at your certification body's discretion.
Transcript Preview
First 800 characters of the transcript
Thank you so much. Um just apologies in that I cannot hear you guys or I can't hear you guys react. So um my name is Emanuel Gera. Uh I am the co-chair of AI Village and uh in my day job I actually work at Google as a chaos containment engineer. And [snorts] so uh today we really wanted to bring this panel together uh in order to discuss the defender dilemma um that being dual use uh AI especially within the context of uh I guess the current and evolving uh landscape. So the reality is that for decades we have been operating as a security community understanding that uh transparency in l or instead of opacity is kind of is the goal. a lot of our reliance on open- source software and just the ability for the security community to poke at things and make sure that we're all on the same page …