China’s eyes, Pakistan’s finger, India’s risk

Xi promised Trump that AI will stay under human control. But a human in the loop means little when the loop runs through Beijing

There is one sentence that hit me. It was spoken last July by India’s Deputy Army Chief, Lt. Gen. Rahul R Singh. Describing Operation Sindoor, he said that when the two sides’ military operations chiefs spoke on the hotline, Pakistan already knew which important Indian weapon system was “primed and ready for action,” and asked India to pull it back. Then came the sentence: “He was getting live inputs from China.”

As a country, we heard this, filed it under “China–Pakistan collusion”, grumbled for a day and moved on. I think that was a serious mistake. That sentence is the most important thing any Indian official has said about nuclear danger in years, and we have not yet understood what it means.

Here is my argument. The world is worried that one day a nuclear power will hand its button to a machine. That is a real fear, and I wrote about it in my last week’s column. But India faces a nearer and stranger danger. Pakistan will not need to hand its button to a machine. It is already handing its eyes to China. And in the age of AI, whoever controls the eyes increasingly controls the decision.

The bargain that kept us alive

Why has nuclear deterrence worked for 80 years? Not because leaders are wise. It has worked because of a brutal bargain: the person who decides to fire is also the person whose own cities will burn. Fear and responsibility sit in the same chair. That is what makes a leader pause.

What Lt Gen Singh described breaks that bargain. In the next India –Pakistan crisis, the eyes may sit in Beijing, the finger in Rawalpindi, and the risk in Delhi, Mumbai and Lahore. China would help shape the picture on which a nuclear-armed army acts, while its own cities stay out of harm’s way. That is influence without exposure. It is the most dangerous kind of power there is, because it carries no reason to be careful.

Beijing played down the General’s claim, and I accept it is the Indian Army’s assessment rather than a proven fact. But nobody should doubt the direction. Around 80 per cent of Pakistan’s imported weapons now come from China. Its jets, missiles, air defences and drones are Chinese. It would be naive to think the data that guides them will be anything else.

Doubt on trusting a borrowed detector

I have spent my working life around scientific instruments, and they teach you one lesson early: never trust a result from a detector you did not build, calibrate or understand. Detectors see ghosts. They mistake noise for signal. That is why it’s difficult to announces a discovery on the strength of somebody else’s black box.

Now think about what AI does to a war room. It does not just pass on photographs. It delivers conclusions: “launcher identified”, “missile armed”, “strike likely”. It delivers them in seconds, in a region where a missile can cross the border in under ten minutes. And it delivers them with a confidence that tired, frightened human beings find almost impossible to resist.

So picture the moment that matters most. A Pakistani general, in the worst hour of his career, looks at a clean, confident screen. He did not build the system. He does not know what data trained it. He cannot know what its makers chose to show him and what they chose to leave out. Yet on that screen may rest a decision about nuclear escalation.

And we must be clear-eyed about the interests of the country that builds the screen. I do not believe Beijing wants a nuclear war in South Asia. No sane state does. But China gains when India is pinned down on its western front. Lt Gen Singh called it killing an enemy “with a borrowed knife”. He also said China was treating the conflict as a “live lab” for its weapons. A laboratory needs test subjects. In that experiment, India is on the other side of the glass.

The hollow heart of “human control”

This is why I found the Washington summit’s language on AI so unsatisfying. Xi Jinping told President Trump that AI must always remain “under human control”. The two sides agreed to set up a channel for AI incidents. Fine words, and a useful step.

But human control of what, exactly? Every major pledge so far, including the one Joe Biden and Xi made in 2024, is about a country’s own trigger. None of them says a word about the targeting data a country pours into a friend’s war. By this definition, a Pakistani general acting on a Chinese machine’s verdict counts as a “human in the loop”.

I say a human in the loop is only as good as the loop. If the loop runs through a third capital, with its own software and its own agenda, then the human at the end of it is not in control in any sense that matters. He is simply the last person to press the button.

Physicists have a name for this kind of trouble. Two bodies moving under each other’s pull are predictable. Add a third, and they are not: a tiny nudge can send the whole system somewhere nobody chose. Washington and Beijing have been working on a two-body problem. South Asia is a three-body problem, and no hotline between two capitals can fix it.

Let us be honest about ourselves

Some will say India is being hypocritical. We, too, receive intelligence from friends. Since 2020, India and the United States have shared geospatial data under the BECA agreement.

That objection deserves an answer, and here it is. The problem is not that friends share information. It is live, automated targeting support flowing into a nuclear crisis from a country that bears none of the risk. India should argue for a rule, not against a country, and it should be willing to live by that rule itself. That is what gives the argument its force. A principle India applies to its own partners is one Beijing will struggle to reject without revealing its intentions.

What India must do

First, India should draw the line out loud. Any country that feeds live targeting data or AI-driven intelligence to a side fighting India should be treated as a participant in that conflict, not a bystander. This is not a threat of war. It is a statement of consequences, and India is not without leverage. Our trade deficit with China crossed $110 billion last year. Beijing cannot be our trading partner in peacetime and our adversary’s eyes in wartime.

Second, India must build its own eyes. The answer to a borrowed eye is a sovereign one. The 52 military surveillance satellites the Government approved in 2024 should be treated as urgent, because in a crisis the side that sees clearly is the side least likely to panic.

Third, India should take the principle to the world. Last year the UN adopted its first resolution on the risks of AI in nuclear command systems. 115 countries voted for it; the nuclear-armed states voted against it or abstained. India, with its no-first-use doctrine, is the nuclear power best placed to break that silence, and to insist that “human control” must cover the whole chain, including sensors, software and foreign intelligence, not just the final button.

Fourth, India should put the question to Beijing directly. China has just offered Washington an AI incident channel. India should ask for the same, along with a plain commitment not to feed live targeting data into any India–Pakistan crisis. If Beijing says yes, South Asia is safer. If it says no, the world will learn exactly what its talk of “human control” is worth.

The next call

Nuclear peace has always depended on one hard truth: those who can start a catastrophe must also share its cost. In South Asia, AI is quietly letting one country step outside that bargain.

There will be another hotline call one day. When a Pakistani general again tells India what he can see, New Delhi must know whose eyes he is using. And Beijing must know that it will answer for what it showed him.

The author is a physicist at the University of North Carolina at Chapel Hill and a contributing opinion writer at The Wall Street Journal; Views presented are personal.