Unbundled

Chapter 1: The Great Unbundling

On a Tuesday morning in late September, a group of Senate staffers huddled in a basement hearing room, debating the precise voltage of a metaphor. The bill before them, the Artificial Intelligence Kill Switch Act of 2026, didn’t literally mandate a switch, of course—there was no physical plug to pull, no circuit breaker to throw. But the language mattered. The word “kill switch” had been chosen deliberately, with all the grim efficacy of a branding consultant who moonlighted as a thriller novelist. It conjured images of a red button under a glass case, a final safeguard against runaway intelligence, a way to stop the machine before the machine stopped us. The staffers argued about trigger thresholds, about who would have the authority to flip the switch—the Secretary of Commerce? A newly created AI Safety Board? A panel of ethicists with tenure and strong opinions?—and about what, exactly, constituted an “emergency grave enough to justify termination.” One staffer, a young woman with a law degree and a visible caffeine tremor, suggested that the bill’s definition of “rogue behavior” was too narrow, that it failed to account for models that didn’t go hostile but merely went weird, drifting into unexpected patterns of output that could destabilize markets or elections without any discernible intent. Another staffer countered that the bill’s language was already too broad, that any sufficiently complex system could be construed as “unpredictable” if you squinted hard enough, and that the kill switch was less a