Australian AI Safety Institute publishes agentic AI risk framework

New research by the Australian AI Safety Institute (AAISI), produced in partnership with the Gradient Institute, has introduced a framework to help organisations, policymakers, and researchers manage risks associated with interactions involving artificial intelligence (AI) agents. 

The report describes how an AI agent uses a large language model (LLM) to plan and execute self-directed actions to achieve human-specified goals. As deployment grows, AI agents will increasingly interact across organisations, partners, suppliers, and open internet environments. These interactions create safety and governance challenges because multi-agent failures can emerge beyond the reach of any single organisation's internal controls.

Building on the Gradient Institute's earlier report on governed LLM-based multi-agent systems, the framework categorises deployments into three governance tiers: singular governance, federated governance, and open environments. It examines specific risk types, failures, and available controls across these tiers, identifying which entities can implement controls and highlighting areas where regulatory or technical gaps remain. The research includes a map that identifies responsible parties and highlights areas of unmanaged risk.

Training Announcement: Freevacy offers a range of independently recognised professional AI governance qualifications and AI Literacy short courses that enable specialist teams to implement robust oversight, benchmark AI governance maturity, and establish a responsible-by-design approach across the entire AI lifecycle. Find out more.