Your unwritten data security control is fear, hesitation, and laziness. AI has none.
Marty Weiss, Ali reza Zadissa
Data protection has relied on syntax and rules at the exits, sitting on top of a permission model everyone knows is too loose. It held because the people with access were afraid of getting caught, hesitated before doing something dumb, and couldn't do much damage before someone noticed. Agents inherit the access and none of the restraint. This talk is about what that breaks, from the permission model that becomes genuinely dangerous overnight, to AI putting an exit in every window. Watching context, behavior, and intent as they happen is what the best programs have always done. With agents in the loop, it's what every program has to do.