Who Holds the Power to Watch?

This is an opinion piece. I’m not telling you what to think. I’m asking you to think.

This is the shorter version of this article (B1-B2 readers). It has shorter sentences and simpler vocabulary. If you prefer more research, more data and more voices from the classroom, there is a C1 for B2 version of this article here with the same argument, the same classroom quotes and the same question at the end.

Here’s the thing about this week. I sat with a security engineer (I’ll call him Marco), and by the end, I agreed with him more than felt comfortable.

Marco watches customer devices for a living. And he thinks the German privacy debate has lost the plot.

“The privacy thing is something we’re discussing too much, too often,” he told me. “We are wasting money, resources, time, energy on some things that is totally the wrong way.”

Of course, one man in one room speaks only for himself. But it stayed with me all week.

The question we discussed was simple to ask and hard to answer: to make AI safe, must we accept being watched — and who decides?

It came from the news. In June, the US government reached into a private AI company and switched off its best model overnight, worldwide, for safety. Europe is still fighting about whether to scan everyone’s private messages to catch a few criminals.

The logic underneath is uncomfortable. To be sure a powerful tool is never misused, someone has to watch how everyone uses it.

So all week I asked myself a different question. Which question are we really arguing about? I don’t think it’s the one on the poster.

Key Vocabulary — What Actually Happened

• dual-use | technology that can be used for good and for harm | doppelter Verwendungszweck

• to surveil | to watch people systematically, usually without their consent | überwachen

• export controls | government rules on who abroad may use a technology | Exportkontrollen

• an error rate | how often a system gets something wrong | Fehlerquote

• proportionate | reasonable compared with the aim; not too much | verhältnismäßig

• function creep | when a tool or power slowly grows beyond its first purpose | schleichende Zweckausweitung

First, the facts. This stopped being abstract in June.

On 12 June 2026, the US government told the AI company Anthropic to block all foreign users from its two best models. The reason was a reported jailbreak: someone had got past the model’s safety limits. Anthropic could not check every user’s nationality fast enough. So it switched the models off for everyone on Earth. Access came back two weeks later, on 1 July, after new safety promises.

Notice the tool they used. The government used export controls, rules built for weapons and microchips, to control a piece of software the whole world was already using. That is a big precedent, whichever side you are on.

The worry is real. Frontier AI is dual-use: the same model that helps design a vaccine can help design a weapon. So the wish to control it makes sense. The question is what you build to do the controlling.

Europe is running the other test. “Chat Control” is a plan to scan private messages for child-abuse material. The European Commission’s own numbers are the best argument against it. The detection tools have an error rate of 13–20%. Nearly half of the reports reaching Germany’s police (the BKA) are not criminal at all. And the confirmed hit rate? 0.000002735% of everything scanned. Read that number again. To catch a few criminals, you would surveil a whole country and find almost nobody. Is that a proportionate way to protect children? On these numbers, no. And (this is my opinion, not a fact) the people writing the rules quietly discuss leaving their own accounts out.

We have seen this pattern before. In the 1990s, the US tried to keep a copy of the key to every private message. It failed by 1996, and strong encryption spread anyway. Tools built to watch rarely stay where they start. That slow spread has a name: function creep. A power built for one purpose drifts to the next.

Germany feels this more than most. The Stasi taught it what a state can do with full information about ordinary lives. In 1983 its highest court said there is “no such thing as a trivial datum.” No piece of data is harmless.

Two strong voices frame the fight. Dario Amodei, head of Anthropic, says “AI-enabled authoritarianism terrifies me”: build the controls before dictators do. Marc Andreessen, the Netscape founder turned investor, warns the opposite: rules written with the big companies protect the big companies.

And it reaches ordinary desks. In 2023, France fined Amazon €32 million for watching warehouse staff too closely, down to a warning for any item scanned in under 1.25 seconds. Not frontier AI. Just a scanner, a stopwatch, and 6,200 people measured to the second.

Key Vocabulary — What the Room Said

• to steelman | to give the strongest, fairest version of a view you disagree with | das stärkste Gegenargument bilden

• to deter | to stop someone acting by making them fear the result | abschrecken

• a works council | an elected group that represents staff in a German company | Betriebsrat

• consent | agreement given freely and knowingly | Einwilligung

• liability | legal responsibility for harm | Haftung

This is where the week earned its keep. Many people in my groups don’t just have opinions about surveillance. They build it, run it, or defend against it for work.

Take Stefan, who works at a network-and-security firm. His rowing club had a vandalism problem, so he installed ten cameras. Motion detection only. Footage deleted after 72 hours. Only the board can see it, behind a two-factor login. He wrote a note for every member explaining why. The cameras are there to deter thieves and to catch the rare bad actor. The objections “evaporated after a ten-minute talk.”

The strongest case for control came from Andreas, at a legal-tech firm, and it was aimed at me. I said determined criminals will get around any rule; they used Google before AI. He agreed, then changed the scale. The truly dangerous people are “hopefully just a handful.” But “if you allow it with AI, then hundreds of people and more are getting this information.” That is the best steelman of the security case all week: AI multiplies the number of capable bad actors.

Then Viktor, an engineer, cut through everything in four words: “Who controls the government?” He made it physical: “government can take control of a self-driving car... they can drive you against the wall.” Dramatic, yes. But it names the real fear. The question is who is watching, and what else they can reach.

Not everyone feared the state most. Katrin works in AI governance. She worried more about what she couldn’t see: “We cannot control what is happening to the information we put in the system.” Then she admitted something honest. She had given her own finances to an AI: “the system knows a lot about me — where I work, how much I earn, where I invest... maybe I’m too open.” Her company still cannot use AI officially, because the works council has not agreed. The blocker is consent, not the technology.

Jonas, a recruiter, drew a clear line. He runs career data through an AI tool, but never “address, phone data, date of birth.” “That’s where I would draw the line.” That line is also where the liability sits.

And the privacy defenders were not paranoid. Bettina named the danger simply: “When I accept small, what is the next step of this? The Germans had a hard past with Stasi — they don’t want to reach this level again.”

Three honest positions. None of them the obvious villain.

Key Vocabulary — Where I Land

• informed consent | agreement given with full understanding of what you agree to | informierte Einwilligung

• to decentralise | to spread power away from one central place | dezentralisieren

• disproportionate | too big or severe compared with the aim | unverhältnismäßig

• a hypocrite | someone who says one thing but does another | Heuchler·in

• to lurk | to wait, hidden, ready to cause harm | lauern

Now I stop reporting and start leaning. I lean, but I am not certain, and you’ll see where I run out of road.

The debate is in the wrong place. Huge effort, almost no result. Scanning a whole country to catch a rounding error is disproportionate, and a little hypocritical too. Please don’t hear “privacy doesn’t matter.” Hear something else: we are spending our fear in the wrong place.

Because here is the real question. It’s not how much we’re watched — it’s who holds the power to watch. Argue about a signposted camera at a boat club, and you miss the machine that should actually scare you.

Stefan’s cameras show the point. Open, signposted, limited, deleted after 72 hours, run by a small club on its own ground: that kind of watching is fine. If you know the camera is there, and why, you behave. A boat club cannot become Big Brother. A state can. So keep the power to watch spread out, with clubs, companies and people, and away from the state. To decentralise the watching is the safeguard.

Private is not automatically safe, though. Meta and the data brokers watch us too. But you can leave Instagram. You cannot leave the state. That is why I trust a thousand small watchers more than one big one — not because companies are angels.

For you, the individual, the deal is informed consent. Understand how a tool uses your data. Accept the parts you can’t remove. That acceptance is what gives you control. Don’t like it? Don’t use it. Don’t care? That’s fine too. The choice is yours.

I know progress is messy. I’ve spoken German badly and confidently for nearly twenty years. If I had waited for perfect grammar before opening my mouth, I would still be silent today. You try, it wobbles, you fix it. Better that than three years of blocking everything to stay safe.

And I keep the hard part in view. Personal data really can be used to build targeted attacks. That’s real. But that, on the other hand, is also life. Most of us are hypocrites about privacy anyway. We protect it when it suits us, then hand everything to a chatbot because it drew a nice pie chart. Katrin admitted it. So would I. So, if you’re honest, would you.

To the reader who carries the Stasi memory: your fear is legitimate, and it is a fear of the state. We want the same thing. Keep the signposted camera. Just make sure the state never holds all of them.

Then the honest ending, where you want me to tie a bow. The lesson’s real question was: who should police the safe use of AI in the future? Government? I don’t want too much government control. The companies? They shouldn’t write their own rules and build a wall around their profits. The AI itself? No. An outside body? I’m not convinced. Four options, and I reject all four. I don’t know who should hold the reins.

And I’ll be honest about myself. I’m not even sure what I would give up to keep these tools. I love what AI has done for my business. I see the huge future and the dark side at the same time. But the dark side has always been there, and most of us just choose to ignore it. It lurks in the background. Always a risk. I use the tools anyway — and if you’re reading this on a phone that knows where you are, so do you.

There’s a companion lesson to this article — the same three positions, built as a free English lesson at B1, B2 and C1: the debate, the vocabulary, and the discussion questions. It runs on The BEBB Method — The Agency Loop, my five-step way of using AI without letting it replace your own thinking. It is my own configuration of established best practice in human-AI collaboration, built in response to Gerlich (2025) on cognitive offloading, and informed by the “AI Sandwich” tradition (Ippolito, 2023) and “AI as Critic” scaffolding (Mollick, 2024). If you teach, you can run the whole thing. Read the companion lesson →

If this was useful, share it with a colleague who’s arguing about the wrong question.

Where do you stand? And what would you actually give up?

Share