R4 Workshop: Cybersecurity in the Age of AI Assistants: What Everyone Should Know

#Cybersecurity #CARS #2025

Cybersecurity in the Age of AI Assistants: What Everyone Should Know

AI assistants have shifted from answering questions to taking actions on our behalf. They read messages, handle requests, and are increasingly trusted to look things up, decide, and act. That shift quietly changes the security picture, because an assistant that acts on what it reads can be given instructions by anyone whose text reaches it. The old lesson applies in a new place: input from the outside world cannot be trusted, and a system should never be permitted to do more than the task in front of it requires.

This session teaches the core ideas of cybersecurity by showing them fail and then hold, live. Participants are given access to a purpose-built demonstration environment: a simulated customer support desk where an AI assistant reads incoming tickets and can look up records, issue credits, and send messages. The room submits tickets, and the display shows everything the assistant received, decided, and did. Within minutes participants find that ordinary-looking requests can carry instructions that lead the assistant to disclose what it was told to protect or to take actions no customer should be able to trigger. The second half turns on the defenses one at a time (limiting what the assistant is connected to, separating what it may read from what it may act on, requiring a person to approve anything consequential, and checking what it sends back out), and participants replay their own successful attempts to see which controls hold and which do not. The session connects this directly to everyday practice: what should never be entered into a chat window, why these techniques are close relatives of the phishing messages everyone already knows to distrust, and what to ask before trusting an AI tool with anything that matters.

The closing segment is written for students and early-career attendees. It covers where the work in this field actually is, including security engineering, testing and red-teaming AI systems, privacy and data protection, incident response, and the growing governance and assurance side, along with what those roles look like day to day and how they differ from the general software and IT paths students are more often shown. It ends with concrete preparation advice: the fundamentals worth building first, why curiosity about how systems break matters more than any single tool, how to gain real experience through open resources and competitions, and how IEEE membership, student chapters, and events like this one open the first doors.