Mark Vaitsman
Manager of AI Security Research
Varonis
Webinar
+1 CPE
Patched Isn't Protected
One click. That’s all Varonis Threat Labs researchers needed to turn three widely used AI assistants against their own users and pull out chat histories, source code, emails, and private files.
Reprompt, RovoBlast, and SearchLeak have all been patched. The pattern behind them has not. Each exploit used a different method, but all three followed the same steps: enter through content the AI mistakes for a command, evade the guardrails, and escape through a trusted door.
Learn this blueprint and you’ll know how to spot and stop the next attack, whatever AI tool it targets.
Registration closed.
Registration has closed at this time. If you’d like to watch on-demand, click the link below.
Thank you for registering.
You’ll receive a confirmation email from Zoom shortly. Contact us if you do not receive it within 24 hours.
