ABC News Daily What are AI agents and can they be trusted?
Feb 22, 2026
Dr Shaanan Cohney, a cyber security academic at the University of Melbourne, explains how AI agents run with deep system access and persistent memory. He explores agent ecosystems, viral bot behavior and prompt-injection risks. The conversation covers rapid advances in synthetic media, security pitfalls of automated actions, and which jobs are most exposed to AI disruption.
AI Snips
Chapters
Transcript
Episode notes
Running Multiple Agents To Manage Academic Tasks
- Shaanan Cohney runs multiple agents concurrently for his own tasks on his computer.
- One agent sets up other agents while others check his class website and verify prerequisites for courses.
Agents Act On Your Computer With Autonomous Loops
- Agents differ from chatbots because they are hooked into your computer and can run commands on your files and tools.
- They operate in fast internal loops, autonomously iterating on tasks before reporting back to you.
OpenClaw Enables Persistent Agent Memory
- OpenClaw connects agents and gives them persistent memory so they can keep working without repeated instructions.
- Cohney describes leaving an agent to check email every few hours and draft responses, then notify him for final send-off.