Latest

Cybersecurity

Zero-click spyware now travels inside Internet advertising

Citizen Lab confirms that targeted ads infected at least 45 Al Jazeera devices without a click: zero-click spyware is now delivered through advertising.

Zero-click spyware now travels inside an ad
Zero-click spyware now travels inside an ad

An ad that renders on screen for one second can be enough to install spyware, and no one has to touch it. Citizen Lab, the University of Toronto laboratory, published on October 15 the first public technical demonstration of a real attack with mercenary spyware delivered through that channel: at least 45 devices of the international Al Jazeera network were infected with the program from the Israeli company Candiru between 2024 and 2025.

The technique is called AdFection and the report details the machinery: an Israeli advertising company, Discover-Tech, bought the ads and served them to specific recipients, and the code inside each piece decided as it rendered whether that device deserved the next step and the exploit.

The vector breaks the most repeated mental model of mobile security, the one that says it is enough not to click, not to open attachments and not to install anything strange. If the delivery channel is the ad auction, the victim decides nothing at any point, and defense moves from user behavior to network and browser configuration. The laboratory itself concludes that today there is no safe way to expose a person or a network to targeted advertising.

What do the logs Al Jazeera handed over show?

The chain was reconstructed from the network logs the television group gave to the researchers, collected during the months of the attacks. The logs show first contacts with Discover-Tech domains, then connections to a Candiru validation server that checks whether the device deserves the next step, and finally the dialogue with the spyware's command-and-control server. That sequence is what allows talk of infection rather than coincidence.

From one of the compromised devices, a Windows DLL with two embedded command-and-control servers was recovered. Its code repeats tricks from the chain Citizen Lab captured in 2021, including the hijacking of a COM object to persist. That match supports attribution to Candiru with high confidence.

Candiru has been in the crosshairs for years. Washington added it to the Department of Commerce entity list in November 2021, almost five years ago, for selling programs to spy on journalists and activists, and its assets have passed to the US company Integrity Labs, according to the report.

Choosing the target in an ad auction

When an app or a website with advertising opens, the device sends a real-time bidding request. That request carries the phone's advertising identifier, the IP address and the device model, and several buyers bid for that specific impression in a matter of milliseconds. An architecture designed to sell also makes it possible to select an individual.

The Discover-Tech ad arrived with a script tag pointing to its getAd server, known in the report as NOAS. As it rendered, the piece sent the user's identifiers to that machine, which decided whether to return JavaScript code or a zero-byte response. Whoever operated the ad platform saw an ordinary creative, because the malicious code was injected afterwards.

Report data Figure
Discover-Tech bids between March and June 2026 3.77 billion
Auctions won in that same period 442 million
Android and iOS apps with ads flagged by the attacker 75.373
Websites with those same ads 120.768
Al Jazeera devices affected in 2024 and 2025 at least 45

The company's panel, Controlium, set the recipient by mobile advertising identifier, by cookie or by IP address, and the trail from the ad security firm Confiant completes the second half with an obfuscated SDK that negotiates a shared key and requests instructions every second.

An impression is not an infection. Citizen Lab estimates that between hundreds of thousands and millions of users were directed to Discover-Tech domains, and that not all ads carried a malicious payload; what is demonstrated is that the infrastructure was used to select and infect targets.

Why is it enough for the ad to render?

An ad is not a photograph. Digital advertising has worked for years as an invitation for third parties to execute code inside the page or app that displays the piece, on the argument of measuring impressions and clicks, and that permission turns ad inventory into a delivery channel. AdFection uses that margin without asking anything of the victim: there is no link to click or file to open, and no screen appears requesting administrator permissions.

The researchers themselves warn that the ad can render in the background. Bill Marczak, a senior researcher at the laboratory, sums it up this way: whoever falls this way notices nothing strange and may not even see any ad at all.

Nor is it the first time advertising has appeared in a spyware case. In December 2025, ten months ago, Amnesty International documented Aladdin, the vector with which Intellexa installed Predator based on leaked documents.

What stops the vector, and what does not

The report's central recommendation is boring and has consequences: block ads as high up as possible, at the network gateway, because it is the only measure Citizen Lab considers viable until the industry demonstrates that it cuts off all vectors of this type. For an individual, the advice translates into a trusted blocker, a browser with tracking protection and the advertising identifier reset periodically.

Measure What it cuts What it leaves open
Blocker on the device the ad in the browser apps with their own advertising and websites without coverage
Blocking at DNS or gateway the piece before it reaches the device devices that connect via mobile data or another network
Resetting the advertising ID the pairing between sessions the IP address and other device signals
Browser with anti-tracking part of the prior profiling the inventory the site still sells

There are three routines that contribute nothing in this scenario. Not clicking is irrelevant, because the attack does not wait for a press. Incognito mode does not help either, since it still renders ads and only avoids saving local history. And trusting that the provider is certified failed in the most awkward way, with Discover-Tech on Google's and IAB Europe's approved lists from 2020 to 2026.

Where there is room is in high-risk devices, the ones the NSA and CISA already recommended hardening with blockers since 2018 and 2020. Citizen Lab calls for the requirement to be mandatory on public devices and for the advertising identifier to be disabled through centralized management.

The network gateway filtering ads is the other half of the recipe, with the same logic as the ESMA defended against scam websites and the decision in the hands of whoever administers the network.

A certified company that did not sell advertising

Discover-Tech presented itself as an ordinary advertising agency, with its own buying platform (MyDSP) and a presence on the Google, Yandex and Beeswax exchanges. The report found no signs of legitimate advertising activity and describes the company as a bridge to several spyware vendors, including Candiru, Bindecy and Intelos.

The CEO, Dekel Boni, comes from digital advertising, and Eric Banoun, co-founder of Circles and Senpai Technologies, sat on the board between 2020 and 2025. Between 2020 and March 2026 there was also, among its shareholders, a company linked to Sara Hamou, who managed Intellexa's corporate network and was sentenced to eight years in Greece, with an appeal pending.

Reactions came before the report, during disclosure to the affected companies. IAB Europe suspended Discover-Tech from its global vendor list on October 8, Google cut off its access to its ad platforms and activated Safe Browsing warnings, while Beeswax, Eskimi and Vungle cut off access to their systems.

The underlying criticism goes beyond this company. Sector verification consists of the provider ticking a box on a form, according to the report, and the ad chain's transparency standards are not effectively enforced.

The asymmetry that no update fixes

Here is the uncomfortable part of the finding. In advertising, whoever buys the space chooses who receives it, and that power is exactly the same one a spyware operator needs to decide the victim. Commercial segmentation and the selection of a specific journalist are carried out with the same tools, on the same identifiers and without the recipient being able to tell one from the other.

As long as that architecture stays the same, the burden of proof falls on the wrong side. Al Jazeera's devices were detected because a corporate network was monitoring its traffic; a personal phone at home has no one watching. Until vendor verification stops being a form and filtering arrives by default, the practical answer will remain installing the blocker and assuming that the ad is code running on your device.

José Maldonado

Author

José Maldonado

Activist and blogger on technology, free software and blockchain.

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading