Existing efforts to regulate AI share a limitation: they’re built around intended use, not the messy, creative and often unintended ways it is actually being used.