Proof of work is broken

There's a concept in Bitcoin called proof of work. To add a page to the ledger, a computer has to grind through an enormous amount of calculation, and the result carries proof that it did. Nobody can cheaply rewrite the ledger, because rewriting it would mean redoing all of that work faster than everyone else combined. The clever part is that the proof is hard to make and easy to check.

The idea is older than Bitcoin. Two computer scientists, Cynthia Dwork and Moni Naor, proposed it in 1992 to fight junk mail. Email had made sending a message nearly free, so they suggested making each one cost a little computing effort: nothing to someone sending a few letters, a lot to someone sending millions. Hashcash put that into practice in 1997, and the Bitcoin paper says plainly that it's built on Hashcash.

We use a heuristic built on proof of work in finding what's real in the real world. We just never called it that. And there's one difference that matters: Bitcoin's proof can be checked. Ours never could. We looked at how finished something was and guessed how much work went into it. The polish was the receipt. Now the receipts print themselves.

Text, pictures, video. They can all be faked, easily and well enough that there's no telling. You can see it with Jean Philanthrope: two of the newest AIs, a subreddit built for exactly this question, and the detectors couldn't settle one dancing man between them.

It goes well past single pictures. Any publication you're not intimately familiar with could be fabricated, whole, by the tens or the hundreds, in just a few minutes, with almost no human attention. That isn't a thought experiment. NewsGuard, a company that rates news sites, had counted 3,749 of them by June: sites mostly written by AI, published without much human oversight, and presented so an ordinary reader would assume there are journalists behind them. They get names like “Times Business News.” And those are only the ones they've found.

There are some things, like WHOIS records, that can establish a website has existed for a while. But those are limited, and under threat as well. The owner's details in them have mostly been hidden for privacy since 2018, and WHOIS itself was retired in January 2025 for a newer system. And a domain's age tells you how old the name is, not how long its current owner has had it. Old names get snapped up the moment they lapse, and bought and sold.

I saw a recent report, one that was amplified by the Department of Homeland Security, saying that there was a mass of migrants headed to the border. How do we establish that's true in today's reality?

And if we do establish there are some, how do we know the number?

I asked Claude to run it down. This is the trail that came back:

As best I can tell, it's real. A wire service reported it leaving and arriving, photographers were at the border, and the Border Patrol acknowledges it. Those are hard to fake all at once, because they're people and institutions with long track records, not just files. Faking all of them together would take a great deal of work, which is the old heuristic again. But it's the part of it that still holds: the work of fooling many independent witnesses at the same time, not the work of making one convincing video.

The number is another matter. 200 to 400 when it left. More than 600 a week later. “Hundreds,” says the Border Patrol. A “mass” by the time it reached the sites that repost other sites. None of those is necessarily a lie. Counting a crowd on the move is hard, and it always was. But every hop is a chance for the number to drift, and a picture of a crowded road can now be made in seconds to match whichever number someone likes.

And look at what it took to get even that far: an AI doing the legwork, and a list of news organizations I already knew to trust. Most people scrolling past have neither. They have the post, the video, and the feeling.

In a cryptocurrency, if proof of work is broken and there's no switch to some other system of verification, then it is no longer a functioning system.

Anyone could write anything into the ledger, and the ledger would mean nothing. I think that's roughly our situation now, minus the switch. The heuristic is broken. The other system isn't built yet.

It can be done. Ethereum, the second-biggest cryptocurrency, switched off proof of work in September 2022 and moved to another way of establishing trust, proof of stake: instead of proving you burned electricity, you put up something you would lose if you cheated. It kept running. Email went through its own version of this. Proof of work never really caught on there, and what saved the inbox instead was mostly filtering, reputation, and proving who sent a message. That's part of why the emails from this site reach you and not your spam folder: the big mail providers now require senders to prove they are who they say they are.

For the rest of us, I think the other system looks something like this:

None of these is free, and none is as easy as the old feeling. That's the point. The feeling was free because a faker would have had to pay. Now that faking is free, the checking has to cost us something instead, a little, every time it matters.

There's a trap on the far side of all this, and I want to name it, because I could walk you right into it. If nothing can be trusted, then nothing has to be believed. Two law professors, Bobby Chesney and Danielle Citron, named it the liar's dividend: once everyone knows video can be faked, anyone caught on real video can just say it's fake.

Hannah Arendt saw the shape of it long before any of this. In an interview in the 1970s she said: “If everybody always lies to you, the consequence is not that you believe the lies, but rather that nobody believes anything any longer.”

The caravan is real. Checking is harder than it used to be, not impossible. The heuristic is broken; the question still has an answer.