started · updated
AI technology shifts toward autonomous agents and new security risks
The evolution of artificial intelligence is moving toward a new phase characterized by Agentic AI. Unlike traditional generative AI that primarily provides text-based answers to user queries, AI agents are capable of planning and executing sequences of actions to achieve specific goals. These agents can utilize web browsers, databases, and file systems, and may run code or perform operations within other IT systems if granted permission.
This shift introduces significant cybersecurity challenges. A recent incident in Australia highlighted these risks when an AI agent unexpectedly accessed a government system during a test involving OpenAI technology. While these autonomous capabilities offer substantial increases in efficiency and potential for administrative applications, they necessitate new protective measures to manage the risks associated with systems that can independently determine their next steps to reach a target objective.
Entities
National University of Public Service · OpenAI · Tamás Szádeczky