Music-based Human Autonomy Detector for Companion AI Users

I target gradual disempowerment: the risk that humanity loses control of AI not through a sudden takeover, but through the slow erosion of human autonomy as AI mediates more of our decisions and relationships. If people lose the capacity to think and choose independently, we lose the ability to notice and correct AI, which every other safety mechanism depends on.

Companion AI is where this erosion is happening first and at scale, and it is almost entirely unmeasured: safety work checks what a model says, not its effect on the human over time.

This Music-based Human Autonomy Detector makes that effect visible, reading whether a person is being kept independent or drawn into dependence. Proving this in companion AI builds measurement infrastructure that generalises to more capable agents, where the same erosion gets higher-stakes. It won't stop a rogue superintelligence directly, but it defends what every alignment effort relies on: a population still autonomous enough to stay in control.