Towards Agentic AI Governance: A Preliminary Assessment

Artificial intelligence is shifting from reactive generative systems—which wait for human prompts to create content—to agentic AI. These systems are designed to perceive their environment, reason through complex problems, and autonomously execute multi-step tasks to achieve user-defined goals with minimal human oversight. As adoption rates rise, agentic AI is increasingly integrated into sectors like finance, healthcare, and urban planning, promising significant productivity gains but introducing novel ethical and safety challenges.

While existing AI governance frameworks and data privacy laws (such as the GDPR) provide a baseline, they are often insufficient for the unique risks posed by agentic systems. The authors highlight that agentic AI introduces specific concerns, including unpredictable behavior, agency problems regarding accountability, and the potential for algorithmic injustice. Despite these risks, the paper finds that most global regulatory bodies, including the EU and the US, have yet to establish specific frameworks for agentic AI. The authors identify Singapore’s 2026 Model AI Governance Framework as a rare, early exception in this regulatory landscape.

Through a systematic review of 21 peer-reviewed studies, the authors identify a "definition syndrome" where the lack of a standardized classification for agentic AI hinders effective policy development. The paper argues that governance must move beyond general AI principles to address the specific "agentic loop"—the iterative cycle of perception, reasoning, action, and learning. The authors propose that future governance must focus on defining clear accountability structures, ensuring value alignment, and managing the risks associated with autonomous goal-pursuit.