This week India's Supreme Court was handed a chance to draw one of the most consequential lines in modern governance, and declined to pick up the pen. An advocate, Narendra Kumar Goswami, had asked for binding guidelines on how the state may use "high-risk" AI, the algorithms that increasingly help decide who receives welfare, who gets watched, and which posts get taken down. Chief Justice Surya Kant's bench refused. "It is a highly technical issue. We are not the experts. And this is a policy domain," he said, sending the petitioner back to the government with his own complaint reclassified as a mere suggestion. The court did note, pointedly, that it had already written comprehensive rules for AI use inside the judiciary itself.
That last detail is the tell. Faced with AI in its own house, the court acted; faced with AI in everyone else's, it deferred. And it is not alone in preferring that someone else hold the pen. Across several countries this month, the institutions best placed to set the boundary between what humans must do and what machines may do all found reasons to look away, even as the boundary was being crossed in front of them.
An essay from the American Enterprise Institute argues the line is not really missing, just unwritten as AI policy. Responding to Mark Zuckerberg's fantasy of everyone owning a "superintelligent lawyer," the author notes that the law already reserves enormous territory for humans. Roughly a fifth of American jobs require an occupational licence a machine cannot hold. Freight trains must carry two human crew; a doctor, not a diagnostic system, must issue the final diagnosis; a pilot must occupy the seat even when software flies the plane. Fresh statutes keep adding to the pile, from Rhode Island mandating staffed checkouts to seven states requiring a human in the loop on insurance decisions. Human privilege is written into the code of the economy one clause at a time, mostly by accident.
Where no one is drawing the line, the vacuum shows. A Washington Post investigation, summarised by The New Republic, found members of Congress and their staff using chatbots to draft speeches, press releases, and even bill amendments, frequently with no idea what the rules are. Congress does have rules: no deepfakes, no constituent data, and humans must finalise legislation. But enforcement rests on staffers reporting one another, and most of those interviewed did not know the rules existed. The gap surfaced comically this summer when a proposed defence-bill amendment reached the floor with the phrase "Claude responded:" still embedded in its text.
The professions are trying to hold their own line, with mixed results. In an August policy paper, the New York City Bar Association concluded that AI may assist legal work but can never substitute for professional judgment, stressing that a lawyer's duty to verify authority is personal and cannot be handed to a machine, a lesson taught the hard way by attorneys sanctioned for filing hallucinated case citations. Sometimes the line is drawn on the spot: an Australian tribunal last week threw out a worker's discrimination claim after catching him reading AI-prepared answers he had stashed on the floor of the witness box.
Put together, the week reads like a system improvising. The underlying doctrine is real. Humans must still act, decide, certify, and answer for the outcome. But nobody wants to be the one who codifies it for AI in general. Courts call it policy, legislatures call it technical, and in the gap the technology keeps advancing into roles no one has explicitly authorised or forbidden. The question is no longer whether AI will be governed. It is whether the lines get drawn deliberately, or one embarrassing incident at a time.