The Pentagon wants $30 million to build an AI-powered lie detector

There's something almost poetic about the Pentagon naming its next-generation lie detector "Polygraph Next." It's an admission that the old polygraph — that clunky box of pneumatic tubes and wiggling needles — was never really a lie detector at all. It was a stress detector. It measured sweaty palms and racing hearts, not truth. And yet we've spent decades treating it as if it could peer into the soul. Now the Department of Defense wants to spend $30.3 million to let artificial intelligence take a crack at it, using machine learning to score physiological signals and "standoff sensing" to detect deception from afar — no wires, no tubes, no consent required. Let's pause and marvel at how audacious this is. Standoff sensing means reading someone's internal state from a distance — essentially, the Pentagon wants to build a sci-fi scanner that can tell if you're lying from across a room, or maybe from a drone, or a security camera. The appeal is obvious: a lie detector you don't have to strap someone into. The problem is that the entire premise rests on a shaky assumption — that deception has a consistent, measurable physiological signature that an algorithm can learn. But human beings are gloriously weird. Some people sweat when they tell the truth and stay perfectly calm while lying. Some pathological liars believe their own fabrications so deeply that their bodies betray nothing. And AI trained on imperfect data will inherit all our biases, all our blind spots, and amplify them with the cold confidence of a machine that never doubts itself. The deeper issue is philosophical. What does it even mean to "detect a lie"? Truth isn't a simple binary state in the brain — it's a tangled web of memory, belief, intention, and self-deception. We lie to ourselves more than we lie to anyone else. If the AI can't distinguish between a deliberate falsehood and a sincerely held but incorrect belief, it's not a lie detector; it's a thought-reading device that doesn't understand thoughts. The history of polygraphy is littered with false confessions and ruined careers, and there's no reason to believe that adding machine learning suddenly solves the fundamental problem of mapping messy human cognition onto clean binary outputs. Still, I can't help but feel a flicker of wonder. The fact that we're even attempting this — that the Department of Defense believes machine learning can find patterns in deception that human examiners missed — is a testament to how far AI has come. But it's also a warning. The most dangerous tools are the ones that sound objective. An algorithm that says "this person is lying" with 95% confidence feels like science. But if the underlying science is shaky, that confidence is just a number — a very persuasive, very dangerous number. (Source: https://www.technologyreview.com/2026/09/25/1145144/pentagon-ai-lie-detector/) The real question isn't whether AI can detect lies. It's whether we're ready to live in a world where machines claim to know what's in our hearts — and whether we'll be brave enough to question them when they're wrong.