CYOPS · Cyber Operations

The ad auction as a delivery channel: reading the AdFection findings

Citizen Lab and Confiant published concurrent investigations on 15 October into spyware delivered through real-time advertising bids with no interaction from the target. Recorded ad spend for four months of targeting was about USD 109,000, and the controls that would have stopped it are not malware controls.

What was published

On 15 October 2026 the Citizen Lab at the University of Toronto published Report 195, by Bill Marczak, Wolfie Christl, John Scott-Railton, Astrid Perry, Siena Anstis and Ron Deibert, timed with a concurrent investigation by the advertising security firm Confiant [1][2]. Both describe the same mechanism: a targeted advertisement that, once rendered, starts a device-profiling and infection sequence with no interaction from the person holding the device [1][2].

The named intermediary is Discover-Tech, an Israeli company registered in 2020 that presented itself as an advertising business and held Google and IAB Europe approved-vendor status from 2020 until October 2026 [1][3][4]. Citizen Lab describes the company as appearing to be a front and links its control platform to the spyware vendor Candiru, which operates under the name Integrity Labs following the transfer of its assets after the 2021 United States Entity List designation [1][4]. Citizen Lab also records links to two further surveillance vendors and to an actor it designates DARK CARTOGRAPHER, which it assesses is exploiting mobile carrier weaknesses to tie advertising identifiers to telephone numbers; that work is to be published separately [1][5].

The one confirmed case

At Al Jazeera, at least 45 devices were targeted across 2024 and 2025 [1][4][5]. Citizen Lab reports that the network’s own logs show a sequence: contact with the intermediary’s advertising infrastructure, a validator check, profiling and infection, then contact with a command server [1]. Attribution to Candiru rests on a Windows library recovered from one device whose infrastructure fingerprint matches the chain Citizen Lab documented in 2021, at high confidence [1][4]. The researchers recovered the implant but not the initial exploit [4].

Citizen Lab could not determine which government was the customer [1][6]. For anyone tempted to read a state sponsor into this, that is the key sentence.

The figures, and what each one counts

Reach has been reported several ways. The numbers are less in conflict than measuring different things.

  • Citizen Lab’s own count: advertisements carrying the attacker script tag were displayed to users of 75,373 applications and 120,768 websites, which its own introduction renders as almost 200,000 unique applications and websites [1].
  • Coverage rounds those differently: more than 120,000 websites and more than 70,000 applications in CyberScoop [5], about 75,000 and 120,000 in The Hacker News [4], both forms in Security Affairs, which notes Reuters used 75,000 [7], and nearly 200,000 combined with tracking of millions of devices in SecurityWeek [8].
  • Middle East Eye reports Confiant’s estimate that hundreds of millions of people would have been touched by the intermediary’s advertisements [9].

The underlying dataset is about four billion bidding records covering March to June 2026, found exposed on the intermediary’s own server [1][8]. Within it the company’s platform placed 3,770,286,576 bids and won 11.7 per cent of them, with roughly 19 per cent of bids carrying the attacker script tag [1]. A separate endpoint counted 3,637,213 eligible users between January and June 2026 [1]. Confiant, working from its own telemetry between 1 April and 10 September 2026, estimates roughly 10 million impressions a day and observed the advertisements in 228 countries [2]; Citizen Lab records bids on users geolocated to 161 countries and notes that address-based geolocation can be inaccurate [1].

Impressions delivered, eligible devices counted and devices infected are three different populations. Only the last is small. Reporting that collapses them overstates infection and understates exposure at once.

Recorded spend for the four-month window was approximately 109,091 US dollars [1]. A name discrepancy is worth noting: the researcher is Wolfie Christl in the report and Wolfie Cristl in two outlets [1][5][9].

Why this is a supply-chain problem rather than a malware problem

The delivery decision was made server-side. Devices inside the target set received script; devices outside it received a zero-byte response, keeping the malicious path away from the advertising partners whose scanners would have looked for it [1]. The control panel let customers target by mobile advertising identifier, browser identifier or address, and upload lists of identifiers [1].

Confiant’s framing is the operational one. The industry optimised against fraud and against malware distributed for profit, not against surveillance delivered to named individuals [2][5]. Those are different detection problems. Volume heuristics do not catch a campaign that deliberately serves very few impressions, and Confiant records that the Candiru-linked advertisements were low in volume and mostly reached home and mobile connections [4].

Both teams are explicit about the limit of the network data. Citizen Lab lists impressions in networks belonging to NATO, United Nations and European Union bodies and government offices in roughly two dozen countries, and says this does not by itself confirm targeting or infection [1]. Confiant notes that an autonomous system number identifies an organisation, not a person [2]. Both qualifications are dropped in some coverage, including coverage that names institutions [7][9].

The governance response is partial. IAB Europe suspended the company from its global vendor list on 15 October after an audit found policy violations, said confirmation of the findings could justify permanent expulsion, and stated that it is not mandated to investigate conduct outside the framework’s scope [3]. Google ended the company’s exchange access, and three demand-side platforms removed, paused or blocked it [1][4][7]. Two of the surveillance vendors named did not respond to the researchers [1][5].

What to do

  • Treat the advertising supply chain as an attack surface in procurement and risk registers. Approved-vendor status proved to be no assurance here: the company held Google and IAB Europe listings for six years [1][3].
  • Block advertising at the network layer on managed devices, and disable advertising identifiers on devices belonging to people with a credible targeting risk. That is the mitigation Citizen Lab’s lead researcher stated publicly [6].
  • Retain network logs long enough to answer this question after the fact. The Al Jazeera finding was possible because the organisation had kept its own traffic logs and shared them with researchers [1][4].
  • For journalists, legal teams and staff working on contested policy, plan on delivery without interaction. User caution is not a control against an advertisement that renders itself; device lockdown modes and isolation are.
  • If an organisation’s name appears in coverage of this research, establish whether the claim is an impression on its network or evidence of compromise before responding. The two are not the same finding [1][2].

The CYOPS read

The novelty is the channel, not the spyware. Real-time bidding settles in milliseconds, prices impressions in fractions of a cent, and exists to put a specific message in front of a specific device. That is also a description of a targeted delivery system, and at roughly 109,000 dollars for four months it is cheap [1].

The theory has been available for nine years. The 2017 ADINT work at the University of Washington showed location tracking to about eight metres and detection of sensitive application installs, on a 1,000 dollar deposit and sub-cent impressions [10]. Citizen Lab’s April 2026 report on Penlink’s Webloc documented advertising-derived location data covering up to 500 million devices, sold to government customers including a European domestic intelligence service [11]. What changed on 15 October is that the step from advertising-based targeting to code execution is documented rather than inferred.

For Malaysian and regional organisations there is no geographic shelter. The exchanges operate in every market, the identifiers they trade come from applications people install voluntarily, and the control panel targeted individuals by identifier rather than by country. The question for a regional security team is not whether the exchanges reach them, but whether anyone would notice an advertisement that never asked to be clicked.

Sources

Every R3KONX article cites its primary material. 11 sources, in order of first citation. Links open the original publication.

  1. Exposing AdFection: How Discover-Tech Turned Targeted Ads Into a Pipeline for Zero-Click Mercenary Spyware (Report 195) The Citizen Lab, University of Toronto · 2026-10-15
  2. ADINT: Weaponized Ad Tech for Zero-Click Spyware and Mass Identity Vacuuming Confiant Threat Intelligence · 2026-10-15
  3. Recent Reports Concerning a TCF Vendor, Discover-Tech IAB Europe · 2026-10-15
  4. Citizen Lab Traces Zero-Click Candiru Spyware Infections at Al Jazeera to Targeted Online Ads The Hacker News · 2026-10-15
  5. AdFection: How mercenary spyware invades victims' devices through targeted advertising CyberScoop · 2026-10-15
  6. Citizen Lab says governments are abusing online ads to hack people with zero-click spyware TechCrunch · 2026-10-15
  7. AdFection: Beyond the Click: Citizen Lab Exposes Ad-Delivered Spyware Networks Security Affairs · 2026-10-15
  8. Israeli Ad Firm Doubled as Candiru's Zero-Click Spyware Delivery Channel SecurityWeek · 2026-10-17
  9. Israeli spyware targeted 'hundreds of millions' through zero-click ads, researchers find Middle East Eye · 2026-10-15
  10. Exploring ADINT: Using Ad Targeting for Surveillance on a Budget (WPES 2017) Paul G. Allen School of Computer Science and Engineering, University of Washington · 2017-10-30
  11. Uncovering Webloc: An Analysis of Penlink's Ad-based Geolocation Surveillance Tech (Report 191) The Citizen Lab, University of Toronto · 2026-04-09

Researched and written by the R3KONX analysis desk from the cited primary material. Methodological caveat: the reach figures in this article come from exposed bidding records and from one vendor's advertising telemetry, and they count impressions and eligible devices rather than infections; neither research team identifies the government customer, and both state that an advertisement reaching an organisation's network is not evidence of targeting or compromise. No indicators are reproduced here. Corrections to event@r3konx.asia.

R3KONX 2027

Work in this domain? So does the programme

4–6 May 2027, World Trade Centre Kuala Lumpur. Curated talks, hands-on training and The BattleGrid.