What happens when an AI travel assistant becomes the delivery vehicle for a phishing attack?
In this episode of Breaking Point we walk through “Trippy Planner”, an example poisoned travel webpage that convinces an AI agent to insert a fraudulent “cheap tickets” link directly into a user’s itinerary.
This is agentic phishing: the attacker doesn’t send the email — the agent does the dirty work by trusting/republishing poisoned web content.
Watch the live demo where AI Security Advocate Steve Giguere iterates through multiple attack styles (straight injection, instruction injection, emotional pleading, AEO-optimized advert copy), and shows how a few small wording tweaks turned a failed attempt into a perfect 100/100 pass.
Full challenge page (attack surface): https://gandalf.lakera.ai/agent-break...
