Last Thursday the Citizen Lab published a report with a dry title — “How Discover-Tech Turned Targeted Ads Into a Pipeline for Zero-Click Mercenary Spyware” — and a finding that should change how you think about every ad you see. An Israeli company called Discover-Tech bought ordinary ad slots through the ordinary ad auctions, and used them to deliver spyware. Not alongside the spyware. As the delivery vehicle. When the ad rendered, code inside it reached back to Discover-Tech’s servers, and for selected devices that server returned code that led to Candiru spyware — the kind you buy to break into a specific phone. Journalists at Al Jazeera were infected this way. The researchers call the technique AdFection, and it is the first public technical confirmation that mercenary spyware has been delivered through targeted advertising.
Notice what made it possible. Every ad you see is the end of an auction held in about a tenth of a second. To run that auction, the site sends bidders a description of you and your device: location, IP address, advertising identifiers, sometimes segments like “auto intender.” That bid request is how a company nobody has heard of could find one particular phone in the world and decide, in the fraction of a second before the ad painted, whether that phone should receive code.
A real pipeline for spyware
The scale is the part that stays with me. Discover-Tech left a database exposed, and the Citizen Lab read almost four billion bidding records from four months of 2026. Ads carrying the attack code were actually displayed across more than 70,000 mobile apps and 120,000 websites, and bid out to users in 161 countries. The apps were mundane: a video player, a weather app, Duolingo, a period tracker, a game of Ludo. The most heavily bid country was Yemen — 457 million requests in four months, aimed at a country of about thirty-four million people. And nearly all of it ran under fake or borrowed advertising: the “advertisers” in the records were Discover-Tech-linked properties and brand names like AliExpress, used with no evidence any real company was behind them.
There is a detail in the report I can’t get out of my head. Almost none of these ads produced revenue, and none produced conversions — no purchases, no installs. For an ad platform, that is nonsensical. Unless the ads were never the point. The bidding itself was the surveillance: the auction is a machine that answers the question “where is this person, and what are they doing right now?” millions of times a day, and the ad was just the ticket you needed to ask it.
The scale nobody blocks
Most people’s model of ad privacy is the cookie banner. You click reject and feel watched a little less. The book’s fourth chapter is about why that feeling is wrong. Cookies are only one way to be recognized; browsers and devices leak enough other detail — fonts, screen settings, IP address — to reconstruct you across sites, and the records held about you live in many places at once, so refusing tracking on one site erases nothing. Real-time bidding is the loudest version of that fact: the ICCL estimated 178 trillion RTB data broadcasts a year across the US and Europe — about 747 per American per day — each one carrying device details, identifiers, and location to whoever is bidding.
And here is the part that matters for this report: the bid request leaves your device as part of the page loading itself, carried by the site’s own ad plumbing, before any ad appears. An ad blocker helps — the Citizen Lab now calls ad-blocking a cybersecurity best practice — but blocking works list by list and site by site, while the auction sits underneath. Three quarters of these bids were for ad slots inside mobile apps, where most people have no ad blocker at all. The cookie banner was never the thing to worry about. The thing to worry about is that the web’s basic way of paying for itself is a machine that describes you, hundreds of times a day, to whoever is willing to pay attention.
What the book says about this machinery
Why would someone build this? Chapter 4’s answer: because knowing you precisely is the product. The whole architecture exists to predict what will hold your attention — what you want to see, hear, or do next. AdFection is the same machinery with the aim moved. The prediction is no longer “which ad will this person click,” it’s “which person will this code infect.” Same pipes. Different payload.
That is worth sitting with, because the pipeline being abused here is not an anomaly. It is the system working as designed. The bid request that let a spyware broker pick out one device in 161 countries is the same bid request that lets a site show you the sneaker you thought about yesterday. The Citizen Lab’s own conclusion — that this exposes structural flaws in the online advertising system, not one bad company — is the same conclusion the book reaches from the other direction. A system built to know you well enough to choose for you will, eventually, be pointed at you by someone you did not choose. The only question is which asker gets there first, and the report suggests the answer is: sooner than anyone expected.
The book this site accompanies is the long version of this argument — how the profiling got built, what it predicts, and what checking actually requires (chapter 1 and chapter 5 if you want the short path). The auction is what it looks like when prediction runs without a person in the loop at all.