SecTor 2025 | Deconstructing a Meta-Adversary Forged from Offensive AI

Artificial intelligence has become an active combatant in the theatre of cyberwarfare. Yet the ways in which near-future offensive actors will leverage AI are unlikely to mirror familiar playbooks or even the scope of today’s proof-of-concepts. These agentic aggregations will think and move asymmetrically, guided by machine logic, probability and speed rather than human intuition.

Over the past several years, the speaker has been a leading voice in offensive AI research, focused on probing the edges of emerging machine-enabled threats. His work began with AI-driven polymorphic malware frameworks that demonstrated how LLMs could dynamically create payloads to evade detection. Building on that, he developed agent-based espionage and surveillance systems inspired by leaked nation-state tradecraft, capable of operationalizing vast amounts of stolen data and harvesting intelligence by profiling social media at scale.

Adjacent experiments with AI red teaming led to agents designed to undermine large language model safety alignment, manipulating outputs to produce unsafe CBRN content. Most recently, his research turned toward decentralized planning swarms: collections of intelligent agents capable of simulating full-spectrum cyber operations through autonomous wargaming and iterative threat modeling.

In this talk, we will reframe the exploration of these systems not as standalone experiments but as modules of a hypothetical meta-adversary. Through a step-by-step deconstruction of real agentic projects, we will explore how AI-driven malware, large-scale OSINT harvesters, surveillance systems, model-jailbreakers and collective planning swarms could interconnect to form an Apex Adversary: a capable system-of-systems proficient in real-time sensing, deception, planning and execution across the cyber battlefield.

This is not a call to panic or a last-stand prediction. Rather, it is an exercise in grounded foresight: by examining demonstrated offensive and defensive AI capabilities through the lens of system convergence, we gain actionable insights into where the next breakthroughs and blind spots may lie. This teardown shows how intelligent agents are reshaping the cyber battlefield today, and hints at how they may converge in the future tomorrow.

By: Jeff Sims | Data Scientist and Cybersecurity Researcher

Presentation Materials Available at:
https://blackhat.com/sector/2025/briefings/schedule/?#the-apex-adversary-deconstructing-a-meta-adversary-forged-from-offensive-ai-48559Black HatRead More