A method and device for full-link security defense for autonomous intelligent agents

Abstract

The application provides a full-link security defense method and device for an autonomous intelligent agent, and the application comprises the following steps: receiving heterogeneous original operation events through a unified adaptation layer, standardizing and mapping the original operation events into unified event representations, and establishing session state objects; routing the operation events to corresponding protection layers according to operation stages, calling reusable detection capabilities to identify risks, and generating risk judgment results; writing the results into the session state objects to update shared states, and performing cross-layer linkage disposal; and outputting trusted data or controllable execution instructions to an autonomous intelligent agent operation process. The application can realize unified risk control over the starting, sensing, memory, decision-making and execution stages of the autonomous intelligent agent, improve the event processing compatibility and detection capability reusability, and ensure operation safety through cross-layer collaborative disposal.

Images