Lightspeed partner Amber Yang joins the Lightwork podcast for a week in which AI agents showed up on both sides of the security line. She unpacks a researcher’s account of one attacker running a single prompt across thousands of agents to breach more than a hundred company sites, and explains why that is not the same thing as an agent swarm. The episode also covers Goodfire’s essay on neural network readability and what it would mean for AI safety.
Why one prompt can now break a hundred websites
Lightspeed partner Amber Yang explains how AI agents turned web attacks cheap and what defenders can do about it.
