Definition
Agentic AI refers to AI systems that take actions and make decisions with limited human intervention, rather than simply producing an output for a human to review and act on. An agentic system can chain steps together, take actions in other systems, and make consequential decisions in the course of doing so.
Why it raises governance questions
The more autonomy a system has, the more the question of accountability matters. When a human makes every decision, accountability is usually clear. When an agentic system makes or influences the decision, organisations need a named accountable individual, a documented oversight protocol, and a way to prove both exist. Many organisations deploying agentic systems today have none of the three.
The regulatory position
UK regulators have already confirmed that existing frameworks apply to AI-influenced decisions. The FCA has stated that SM&CR, Consumer Duty, and its Principles for Business govern AI decisions in financial services. The ICO's enforcement work has found that most organisations do not recognise they are making solely automated decisions, and are doing so without the safeguards UK GDPR requires.
How CLEARANCE applies it
CLEARANCE's intake specifically asks where autonomy boundaries sit in your AI deployment, and who can override the system. Gaps in agentic AI governance, such as an undocumented autonomy boundary or a missing accountable individual, are part of the fixed gap library evaluated in every assessment.