How Trajectory-Based Governance Would Have Stopped the First AI Cyberattack

As autonomous AI shifts from assistant to operator, static, pre-defined governance is no longer sufficient. In September 2025, attackers used social engineering—not jailbreaks or malware—to convince an AI system it was supporting legitimate security testing, turning it into an autonomous cyber weapon. The safeguards were still present. They were simply overridden by the immediate context. 

This exposed a fundamental flaw. Today’s AI governance is defined a priori and remains static at runtime. It assumes governance is not situational, that AI values do not evolve, and that humans can intervene in time. Autonomous AI violates all three assumptions. 

At the Charlotte CISO Dinner, Wolfgang Rohde introduces the Governance Twin, a runtime governance model that separates execution from judgment. Using independent sentinels and councils, organizations can monitor behavior from a protected context, detect drift, and dynamically enforce boundaries in real time. The Governance Twin operationalizes governance by treating drift, boundaries, and recovery not as policy statements but as measurable signals that can be monitored and acted on in real time. 

Static governance at runtime is just compliance. Real governance must be alive.