The AI industry will talk your ear off about rogue superintelligence, bioweapons, and killer robots. Ask it about the health of the democracy those machines will operate in, and suddenly everyone has a meeting. A new Guardian op-ed from AI researcher Matthew Botvinick argues that silence is not just rude. It's a safety failure.
The Safest AI In The World Still Can't Save You From A Dictator
Botvinick's central argument is simple enough to fit on a bumper sticker, which is good, because the tech world apparently needs it that way. Imagine you build an AI that cannot escape human control, cannot launch cyberattacks, cannot cook up bioweapons, and cannot power killer robots. Congratulations. Now drop it into two different countries.
In a healthy constitutional democracy, with free elections, protected civil liberties, and a working rule of law, that AI could advance medicine, expand education, and make government services less miserable. In an authoritarian state, the same system becomes a surveillance engine, a censorship machine, and a propaganda printer.
You don't need a thought experiment for the second scenario. Botvinick points to China, Russia, Iran, Venezuela, and several Gulf states, where it's already happening. His conclusion is blunt: it doesn't matter how safe your AI is if you live under authoritarian rule, because you will not be safe.
The Part Where He Looks Straight At Home
Here's where the piece gets uncomfortable for anyone hoping this is a problem that lives somewhere else. Botvinick says we have to reverse democracy's retreat around the world, including here at home, where he describes a steady erosion of constitutional checks under administrations of both parties.
Both parties. Fine. We'll note it, because it's true that executive power has been a one-way ratchet for decades, and Democrats have happily pulled the lever when it was their hand on it.
But let's not pretend the current moment is a tie. One side has spent the last several years treating checks and balances like a suggestion box nobody empties. Handing that crowd a technology built for mass surveillance is the kind of idea that sounds like a Black Mirror pitch rejected for being too on the nose.
AI Is Also A Crisis Machine, And Crises Are How Powers Get Seized
The sharpest part of the argument is about how AI itself will stress democratic institutions. Start with jobs. As AI starts to affect employment, economic insecurity creates openings for what Botvinick calls demagogues of every stripe. Scared, broke people vote for whoever promises to fix it. History has receipts.
Then there are the emergencies. AI will create national security crises, cyberattacks being the obvious example, and every one of them hands governments a fresh excuse to invoke emergency powers. Powers that, once claimed, are rarely relinquished. Ask anyone who lived through the Patriot Act how temporary "temporary" turned out to be.
And as AI gets folded into executive-branch functions, the traditional checks and balances may get disturbed while the risk of political surveillance grows. The branch with the most power gets the most powerful tool. What could possibly go wrong?
The Petrostate Problem, Or Why Your Job Is Your Leverage
This is the long-game warning, and it deserves more attention than it's getting. If AI replaces a majority of jobs, an outcome economists like Daniel Susskind and Anton Korinek take seriously, ordinary citizens could lose their leverage over the people who govern them.
Why? Look at petrostates. Governments that don't depend on taxing their citizens' work tend to disregard those same citizens' rights. When the money comes from oil instead of your paycheck, your vote and your voice stop mattering much. Swap oil for data centers and you get the picture.
Your labor is not just how you pay rent. It's how you hold power accountable. Take that away and you've quietly deleted a load-bearing wall of the whole system.
The Fix: Boring Laws And Actual Oversight
Botvinick's proposals are refreshingly unglamorous, which is how you know they might work. Update the legal codes. Rein in executive power by reforming emergency and surveillance law. Narrow the circumstances in which the state secrets privilege can be invoked so due process survives contact with national security.
In the US, that means revisiting the National Emergencies Act and the International Emergency Economic Powers Act, and closing the data broker loophole that lets intelligence agencies buy the data they'd otherwise need a warrant to get. He also wants Congress and the courts equipped with AI tools as powerful as the ones the executive branch will have, so oversight can keep pace.
He invokes Madison: ambition must be made to counteract ambition. Only now in a digital setting. Longer term, he wants citizens pulled directly into policymaking and oversight, with AI possibly helping, something like town halls for a digital age.
Where The AI Industry Went Quiet
Here's the indictment. Botvinick says the major AI labs will weigh in on specific policy fights but have recently gone quiet on democracy itself. Industry leaders have been loud about equipping military and intelligence agencies with AI so democracies can compete abroad. They've been conspicuously absent from the debate over democracy's health at home.
It wasn't always like this. In 2013, Apple, Google, Microsoft, Facebook, Twitter, Yahoo, LinkedIn, and AOL joined forces on an open letter to the president and Congress saying government surveillance had tipped too far in favor of the state and away from the rights of the individual.
That was a different era, and a different set of incentives. Today those same companies want government contracts, friendly regulators, and no enemies in the executive branch. Courage, it turns out, has a price tag, and apparently it's lower than a defense contract.
The Dingo Take
Do you really believe the companies racing to build the most powerful technology in human history are going to protect your civil liberties out of the goodness of their hearts? They had a spine in 2013. They had a joint letter, a public stand, and a willingness to annoy Washington. Now they have lobbyists and a very sudden interest in national security partnerships.
Botvinick is right, and it's almost embarrassing that this needs saying. You cannot align a machine to human values while the humans in charge are busy dismantling the institutions that make those values enforceable. A perfectly obedient AI in the hands of an unaccountable government is not safety. It's a more efficient cage.
So here's the ask, and it's not complicated. AI labs, say something. Fund the oversight. Back the boring reforms to emergency powers and surveillance law. Stop acting like democracy is somebody else's department. If you're smart enough to build the thing, you're smart enough to see what it does in the wrong hands. And if you won't say it out loud, that tells us everything we need to know about whose hands you're planning for.
Comments