AI Surveillance: Age Verification Meets Your Chat Logs
From the Trenches
SubscribeSign in
The Cost of Reading Everyone Just Hit Zero<br>Age verification and the log of how you think are converging into one chain. The main thing that ever protected you was price.
Denis Stetskov<br>Jun 25, 2026
Share
On July 8, 2026, Anthropic can ask for your passport. The updated consumer privacy policy adds a category it calls Verification Data: an image of your government ID, the information printed on it, a photo or video of your face, and a facial geometry template. It covers Claude Free, Pro, and Max. It does not touch the API. A vendor named Persona collects the documents and the selfies, processing them on Anthropic’s behalf as a service provider rather than an independent data controller. The company says only a flagged subset of users will ever see the prompt.<br>Persona itself was caught in February with its government dashboard exposed on a public endpoint, the code showing a pipeline of 269 verification checks and three-year retention of faces and IDs.<br>Handing over a face is trivial. What sits behind it is a year of questions you would never put your name to, and reading all of it just turned cheap.<br>Payroll was the ceiling
For most of history, surveillance ran into the same wall: payroll. A Soviet censor could open your letter. He could not open everyone’s letter, because steaming envelopes takes hands and there were never enough hands. The Stasi got closest, running files on a share of the East German population large enough that historians still argue the exact fraction, and it took warehouses of paper and a network of informants the size of an army to do it. The ceiling was human. Reading you cost a person a day, and a person’s day costs money.<br>What protected the ordinary person was the price of attention. You were boring, and boring was expensive to read.<br>AI removed the ceiling. A model reads every message at a cost rounding to zero, sorts them, flags the interesting ones, and never asks for a raise. The thing that made mass reading impossible was arithmetic, and the arithmetic just inverted. Two variables remain: who holds the data, and what they say when the state asks.<br>Refusal is the exception
EU Commissioner Thierry Breton sent Elon Musk a letter in August 2024 warning that a live interview with a US presidential candidate could trigger DSA measures against X. Musk refused. In December 2025 the Commission fined X roughly 120 million euros, and X is challenging it in court, the first legal challenge to a DSA fine. Whether the fine holds, whether X folds quietly later, we do not know yet.<br>The other response leaves no press release. TikTok rewrote its global community guidelines across 2023 and 2024 to satisfy DSA pressure, restricting categories of political speech worldwide, and users were never told that was the reason. It surfaced only because the US House Judiciary Committee subpoenaed the internal documents and published them.<br>Telegram is the cleaner case. A month after French authorities arrested Pavel Durov in August 2024 on charges tied to content on the platform, Telegram changed its terms to hand IP addresses and phone numbers to authorities on valid legal request, up from the 14 requests covering 108 users it had honored earlier that year. Durov framed it as global consistency. I read the timing as a man doing the math after four days in custody, and I will mark that as my read, not a proven quid pro quo. What he was actually shown, and what was actually agreed, never became public. That is the part that should bother you, not the man.<br>The faster route skips the asking and writes the apparatus into law, justified by something nobody can vote against. One country already ran that justification all the way to the end.<br>Same rails, better manners
Russia stopped bothering with manners early. Vladimir Putin signed Federal Law 139-FZ on July 28, 2012, an amendment to the statute on protecting children from harmful information, child abuse imagery, drugs, suicide methods. It created a single national registry of blocked sites under Roskomnadzor, in force that November. By the end of 2013 the Lugovoi law had bolted extrajudicial blocking for extremism and calls to protest onto the same registry, and by 2022 it was the instrument shutting down independent outlets and war coverage. The children were the door the apparatus walked through.<br>The democracies are building the same lever with better manners. The UK Online Safety Act and the EU Digital Services Act are sold as child safety, and so is every age-verification law in the current wave. Strip the packaging and the demand is the same: identify the user, filter what they see, and on order, block. Brussels runs it with judicial review and oversight that Moscow never bothered with. The lever underneath is identical, and the difference is how many press conferences you hold about protecting kids while you install it.<br>I have watched a ban fail
That excuse...