Unlocking Safe Agentic Autonomy through Verification (Keynote)

As the agentic capabilities of frontier models and frameworks advance, a vast array of tasks across software engineering and knowledge work are being delegated to autonomous systems. Yet, a critical barrier remains: a deficit of trust. Issues of misalignment, probabilistic shortcuts, and containment dominate the current conversation. The proverb "Trust, but verify" has never been more relevant. In this keynote, we will explore the necessity of safely unlocking agentic autonomy. We will make the case for investing in formally provable specifications and deterministic verification techniques across academia and industry which will be paramount to realizing the true, safe impact of Agentic AI.