Governing What We Cannot See: Agentic Shadow AI in Cybersecurity

CIGI Papers No. 366

November 18, 2026

AI for security operations is outstripping current cybersecurity governance, and with agentic shadow AI (autonomous and semi-autonomous cybersecurity systems deployed without institutional visibility, authorization or accountability) emerging, the difference is becoming staggering. As AI is ever less confined to controlled research settings and increasingly intertwined in the routine operations of organizations of all kinds, stringent cybersecurity regulation is a top concern.

Abbas Yazdinejad proposes a cooperative AI governance model designed to address this breakdown in institutional governance built on four pillars: institutional visibility mechanisms; human-AI co-decision structures; multi-actor coordination across organizations and borders; and alignment of AI agent behaviour with shared governance goals.

About the Author

Abbas Yazdinejad is a post-doctoral scholar in the Artificial Intelligence and Mathematical Modelling Lab at the University of Toronto.