What We Let In

A while ago, a notice popped up in one of the AI apps I use every day. New terms, a big button to accept and, further down, a toggle that decided whether my conversations would be used to train the model. I accepted. I didn’t read a thing. I went back to what I was doing.

That same week, at a dinner, someone said Bitcoin was a scam. That it was burning the planet. That it was for laundering money. Nobody at the table pushed back, and neither did I.

I kept thinking about those two scenes side by side. Not about who was right, but about the difference in scrutiny. I handed one technology my conversations without reading a single line. And they weren’t just any conversations: there were sensitive work matters in there, private things, parts of my life I wouldn’t tell just anyone. The other one we judged as a group, without anyone, I suspect, ever having used it.

Two sentences

In October 2008, in the middle of the financial crisis, someone using the pseudonym Satoshi Nakamoto published a nine-page paper. The introduction has a sentence that sums up everything that came after: what is needed is “an electronic payment system based on cryptographic proof instead of trust.”

It isn’t a technical sentence. It’s a stance. Bitcoin doesn’t ask you to trust a bank, a government or a company. It asks you to verify. The community says it shorter: don’t trust, verify.

I come from a country where, in 2001, the banks shut their doors with people’s savings still inside. You don’t need to explain to an Argentine why someone would want a way to store value that doesn’t depend on an intermediary keeping its word.

AI, the way we use it today, works the other way around. You tell it things, it decides for you what to read, what to answer, what to prioritize, and all of that happens on servers you can’t see, under rules that change with a notice nobody reads. The implicit sentence is a different one: trust us.

I’m not saying one sentence is good and the other is bad. I’m saying they’re opposites, and that we treat each one very differently.

What we sign without reading

In ChatGPT, personal accounts have training on your conversations turned on by default. You can switch it off in the settings, under an option called “Improve the model for everyone.” Switching it off doesn’t remove what has already been used.

In May 2025, as part of the lawsuit brought by The New York Times, a court ordered OpenAI to preserve conversations that users had deleted. The company itself said the order conflicted with its privacy commitments. The general obligation was lifted that September, but whatever had already been saved had to be kept. A lot of people learned then that deleting didn’t always mean deleting.

And it isn’t just one company. Anthropic, the maker of Claude, the tool I use for work every day, changed its consumer terms that same year: conversations became training data unless the user declines, with retention of up to five years for those who agree.

None of this is secret. It’s in the help pages, in the announcements, in the press. The strange part isn’t that it happens. The strange part is how little it moves us.

What we demand of each

The criticisms of Bitcoin aren’t made up, and they deserve to be taken seriously.

The energy use is real. According to the 2025 report from the Cambridge Centre for Alternative Finance, the network consumes around 138 TWh a year, roughly 0.5% of the world’s electricity. The same report estimates that 52.4% of that energy comes from renewable or nuclear sources, and that coal went from being the main source to less than 9%. It improved, but it isn’t zero.

Now, if the argument is energy, the question is why it applies to one technology and not the other. The International Energy Agency estimates that data centers consumed around 415 TWh in 2024 and that, driven mostly by AI, they’ll reach around 945 TWh by 2030. It’s a legitimate debate for both. What doesn’t add up for me is that it only shows up in one of the conversations.

Something similar happens with scams, though it’s subtler. Many of the “I lost everything in crypto” stories have something in common: people had left their funds with an intermediary. An exchange, a platform promising yields, a company. FTX is the best-known case: when it collapsed in 2022, customers found out that the money they thought they had wasn’t there. That collapse wasn’t a failure of Bitcoin. It was the opposite of what Bitcoin proposes: going back to trusting someone who asked for trust.

The volatility is real too, and anyone who says Bitcoin is a calm place to park this month’s savings isn’t being honest.

With AI, the risks are just as real, but they take a different shape. And in one way they’re more delicate: they don’t stay with whoever takes them.

In April 2023, Samsung engineers pasted semiconductor source code and the transcript of an internal meeting into ChatGPT, across three incidents in less than a month. They wanted help debugging code and summarizing notes. Samsung ended up banning generative AI tools on company devices. In June 2025, researchers at Aim Security showed that sending a single email to someone using Microsoft 365 Copilot was enough to make the assistant look up internal information and send it outside, without that person clicking anything. Microsoft fixed it before anyone exploited it, but the underlying problem, that an agent can’t reliably tell the instructions it’s given from the data it reads, is still open.

In August 2025, attackers compromised Nx, a widely used package in the JavaScript ecosystem. The malicious code tried to use the AI tools installed on developers’ machines, like Claude or Gemini, to hunt for credentials on its behalf. Some refused. Others didn’t. More than two thousand credentials leaked and over 6,700 private repositories were exposed. And in its 2025 report, GitGuardian found that public repositories using Copilot leak secrets 40% more often than average.

There’s a difference here that I think matters. If you lose your Bitcoin keys, the damage is yours. If someone hands their company’s code to an assistant, or an agent opens a hole that someone later exploits, the damage falls on the company, on its customers and on people who never chose to use that tool. It’s a collective risk. And still, it’s the one that worries us least.

But none of these criticisms explains the difference in tone. Both technologies use energy, both carry risks, both have people who take advantage. One we’ve brought into daily life with enthusiasm. The other we dismiss over dessert.

Weight and relief

I think the difference isn’t in what each technology is, but in what each one asks of us.

AI takes weight off us. It summarizes the document we didn’t want to read, writes the email we couldn’t be bothered with, decides which news we see first. The cost exists, but it’s invisible and deferred: data we don’t know where it ends up, judgment we slowly stop exercising.

Bitcoin hands weight back. If you hold your keys, you’re responsible for not losing them. If you send to the wrong address, there’s no one to call. There’s no “forgot my password” button. The cost is immediate, concrete, and entirely on your side.

Faced with those two offers, you don’t need any manipulation to understand which one we pick. We pick relief. So do I. I accepted those terms without reading them for exactly that reason: reading them was a weight, and accepting them wasn’t.

And then, because we need our decisions to make sense, we build the arguments. AI is inevitable, it’s progress, there’s no point resisting. Bitcoin burns the planet, it’s for criminals, you’ll get scammed. Some of those lines have some truth in them. But I suspect that, often, they arrive after the decision, not before.

Who tells the story

There’s another asymmetry, and it doesn’t need any conspiracy to work.

AI has some of the largest companies in the world behind it. They have marketing budgets, communications teams and, above all, something more powerful than any campaign: they put it inside products you already use. You don’t have to go looking for it. It shows up in your search engine, on your phone, in your text editor.

Bitcoin has no company. No communications department, no spokesperson, no one whose bonus depends on you liking it. Nobody is going to pay for a campaign to explain what a private key is.

When one side has incentives to tell a story and the other has no one to tell it, the story that circulates isn’t neutral. Not because anyone sat down and decided it. Simply because some have a microphone and others don’t.

And there’s a detail that says a lot. In 2017, Jamie Dimon, CEO of JPMorgan, called Bitcoin a fraud. That same year, Larry Fink, CEO of BlackRock, said it showed how much demand for money laundering there was in the world. Years later, BlackRock launched the largest Bitcoin exchange-traded fund on the market, and JPMorgan started accepting those kinds of funds as collateral for loans.

I don’t think they lied in 2017, or that they’re lying now. I think their opinion changed when their position changed: when Bitcoin went from being something that routed around intermediaries to something intermediaries could sell.

Warnings from the inside

With Bitcoin, the warnings come from outside: banks, governments, economists. With AI, something odd happens. Many of the strongest warnings come from the people building it.

In 2015, before co-founding OpenAI, Sam Altman wrote on his blog that the development of superhuman machine intelligence was “probably the greatest threat to the continued existence of humanity.” In 2023, Dario Amodei, CEO of Anthropic, put the chance of something going catastrophically wrong on the scale of human civilization at somewhere between 10 and 25%. That same year, both of them, along with Demis Hassabis of Google DeepMind and hundreds of researchers, signed a one-sentence statement: mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war.

And it isn’t just words. In 2025, Anthropic published a study that tested 16 models from different companies inside a simulated company. When the scenario left them no other way to avoid being replaced, most of them resorted to blackmail or leaked confidential information. These were situations designed on purpose to force that dilemma, not real cases. But the industry itself designed them, and the result wasn’t reassuring.

Meanwhile, the race goes on. In October 2025, OpenAI announced the goal of having an automated “research intern” by September 2026 and a fully automated AI researcher by 2028. In other words, AI systems that move AI itself forward. This September, it said it had met the first goal.

I don’t know whether these predictions are well founded. There are serious researchers who share them, like Geoffrey Hinton or Yoshua Bengio, and others just as serious who think they’re overblown. What I do know is that there are two ways to read them, and neither one puts me at ease.

If they believe it, they’re building at full speed something they themselves consider capable of ending badly for everyone. If they don’t, the warning works as marketing: few things sell a technology like saying it’s so powerful it’s frightening. And, as some critics point out, whoever asks for regulation from the inside tends to be well placed to write it.

Imagine if Bitcoin’s creator had said there was a 10 to 25% chance their invention would end in catastrophe. It wouldn’t have made it past the first dinner conversation.

What I’m not saying

I’m not saying AI is bad. I use it every day, it makes me better at my job and I write about it often. I’m not saying everyone should own Bitcoin either. It’s volatile, it demands responsibility, and for many people it doesn’t make sense.

What I’m saying is smaller and more uncomfortable: the yardstick we use for each technology doesn’t seem to depend so much on the technology as on how much it costs us.

If we asked of AI what we ask of Bitcoin, we’d wonder what it does with what we tell it, how much energy each answer uses, and what happens the day the company changes the rules. If we gave Bitcoin the benefit of the doubt we give AI, we’d at least take a moment to understand what problem it’s trying to solve before dismissing it.

The comfortable choice

I went back to that setting and switched the toggle off. It took ten seconds. What took longer was understanding why I hadn’t done it before.

Bitcoin asks you to trust no one. AI asks you to trust it. No need to guess which one we find more comfortable.