Black Hat USA 2026 | AI and the Future of Cyber Defense Panel

A Black Hat panel of security leaders from Luta Security, PwC, OpenAI, and Anthropic agreed that adopting AI for cyber defense is inevitable—organizations must deliberately integrate it now, with governance and identity controls for AI agents, rather than rely on patching 2:07-2:51.

Key Takeaways:
• AI adoption for defense is "inevitable"—the risk mitigation is to turn AI toward the risk and embrace it, rather than fear shadow AI 2:07-2:51
• Security must be built into AI integration from the start—governance, accountability, and monitoring of agents—since bolting security on later is expensive; full autonomy still requires human-in-the-loop for consequential actions 3:51-5:09
• AI adds a "fast-forward button" to the existing attack playbook, so defenders must accelerate too; human-in-the-loop for every defensive action doesn't scale—reserve it for irreversible actions 8:16-9:38
• You can't patch your way out: AI-driven vulnerability discovery is creating exponential volume, so prioritize reducing attack surface, segmentation, and containment planning instead 17:31-20:41
• Agent identity is the most critical control: you can't authorize what you can't identify—inventory agents, maintain audit trails, and use supervisory models for dynamic authorization 25:19-27:55
• For critical infrastructure/OT, AI can help with asset inventory (only ~2 of 10 operators understand their assets), protocol/physics-level visibility, virtual patching, and digital twins—and "air gapped" is not a security strategy 12:56-15:55

Panelists' parting advice: leverage AI for asset inventory and prioritization, red-team continuously with frontier models, don't over-rely on compliance, and don't assume AI replaces human expertise—build the next generation of talent who can direct AI with intent 38:28-40:41.

Sources:

• 2:07 Katie Moussouris: using AI for defense is inevitable; mitigate AI risk with AI
• 3:51 Morgan Adamski: embrace disruption, build governance/security in from the start, human-in-the-loop for consequential actions
• 8:16 OpenAI's Faiz: AI fast-forwards the same attack playbook; defenders must scale with AI
• 17:31 Panel on patching limits: reduce attack surface, segmentation, and basics over patch volume
• 25:19 Agent identity, audit trails, dynamic authorization as top controls
• 12:56 Anthropic's Sergiy on OT/critical infrastructure: AI for asset inventory, visibility, air-gaps are not a solution
• 38:28 Lightning round final advice from all panelists