Glossary
What is agentic AI accountability?
Agentic AI accountability is the ability to answer, for any decision an AI agent influenced, which agent acted, on whose authority, against which policy, with what data, and with what outcome.
§ 01 Definition
The operator owns the outcome.
Agentic AI accountability is the ability to answer, for any decision an AI agent influenced, which agent acted, on whose authority, against which policy, with what data, and with what outcome. The accountability sits with the operator that deployed the agent, not with the model vendor or the harness designer. The model is a component; the harness coordinates components; the operator owns the workflow and the outcome.
What changes with agentic AI is not whether the operator is accountable, but what makes the accountability defensible. For a human-only workflow, the audit trail follows the people; for an agentic workflow, the audit trail has to follow the agents. The substrate is the layer that makes that record producible.
§ 02 Questions
Agentic AI accountability, answered.
Who carries the accountability for an agentic AI decision?
The operator deploying the agent. Model vendors carry product-side obligations under the EU AI Act and other frames, but the institution that put the agent into production is the institution an examiner asks. The accountability does not move because the work was done by a non-human actor.
What evidence does agentic AI accountability presuppose?
A durable, attributable, tamper-evident audit trail of what each agent did, on what data, under which policy, in what sequence, and with what outcome. Without that trail, accountability is a posture; with it, accountability is provable. See audit by construction and agent audit trail.
Which regulatory frames frame accountability for agentic AI?
SR 26-2 points banks to broader risk governance for agentic AI. The EU AI Act places operator obligations on deployers of high-risk AI. DORA holds financial entities accountable for the ICT operations their agents produce. NIST AI RMF 1.0 frames accountability through the Govern and Manage functions. NERC CIP treats accountability for agent actions on Bulk Electric System assets the same as accountability for human actions.
§ 03 Related
Where the term lives.
Regulatory frames
Related terms