AI is about to make software too secure—and that's a problem

Everything is about to "go dark"

At Usenix Security, the author reflects on how AI-powered vulnerability hunting could soon eliminate remotely exploitable bugs in major software. While great for security, this threatens law enforcement and intelligence agencies' ability to hack devices, potentially reigniting demands for backdoors. The author warns this could lead to weakened systems and increased risk from foreign adversaries, urging a careful path forward.