infosecanon
Heather Lawrence is a Senior Data Scientist with the Electricity ISAC and owner of Atropos Insights, specializing in building custom AI solutions and teaching ML with Python. Her expertise in AI security stems from over a decade of engineering experience, including incident response at Block, Inc., security data science at ThreatKey, and service as a U.S. Navy nuclear engineer. She earned her PhD at the University of Colorado, Colorado Springs in Security in 2026 and her research centers on network security, adversarial machine learning, and the security of AI-orchestrated environments. She has presented at several venues, including the IEEE Conference on Communications and Network Security and the DEF CON AI Village, and volunteers with DEF CON.
Session
The last two years represent a paradigm shift from AI as a "chat assistant" to an autonomous agentic workforce. Meanwhile, "vibe hacking" and prompt injection are rapidly evolving into machine-speed warfare driven by autonomous AI agents. As models continue their exponential parameter growth, threat actors are now leveraging agentic connectors for zero-click data exfiltration (AgentFlayer) and utilizing AI-orchestrated espionage campaigns (GTG-1002) that operate at speeds physically impossible for human defenders to counter. Furthermore, "semantic corruption" through disinformation networks, like Pravda, now target the models' internal logic by flooding and poisoning their training and retrieval data.
This evolution is driven by persistent orchestration frameworks like GasTown and RalphWiggum, which move away from ephemeral sessions toward Git-backed work ledgers and the "GUPP" (Execute Immediately) principle. While these systems offer massive productivity gains, "vibe coding" builds software with intent rather than manual review. We have fundamentally expanded the attack surface while we have little visibility into the silent, machine-to-machine, exploitation of the "agent stack." This talk explores how these autonomous architectures have decoupled technical execution from human oversight, requiring a complete rethink of enterprise trust boundaries. [183]