# The Ad Loaded. The Spyware Did Not Need Your Click

> Researchers traced mercenary spyware on Al Jazeera devices to ordinary online ads. Here is how an ad auction became a targeting route, what the evidence proves, and how high-risk people and newsrooms can reduce the exposure.

- **Author:** Kubilay Tunca
- **Published:** 2026-10-17
- **Category:** For Experts
- **Tags:** Targeted Surveillance, Spyware, Advertising Technology, Privacy
- **Canonical URL:** https://cyber-security-in-plain-english.com/post/experts/news/online-ad-can-become-spyware-delivery-route

---

You open a news site. An advert fills a rectangle beside the story. You do not tap it. You do not download anything. The page looks ordinary, so you keep reading.

That quiet moment is now part of a documented spyware case. On 15 October 2026, the Citizen Lab at the University of Toronto published evidence that targeted online ads had been used to infect devices at Al Jazeera with mercenary spyware. At least 45 devices were targeted during 2024 and 2025. The researchers linked the operation with high confidence to spyware made by Candiru, a company now reported to operate as Integrity Labs. The person at the screen did not need to click the ad.

The finding deserves attention without turning every banner into a reason to panic. Citizen Lab documented a focused, expensive operation against a newsroom. Its report does not show that every person served an ad was infected. It shows that the machinery built to find the right consumer at the right moment can also find one journalist, inspect the device, and hand that device to a spyware operator.

For reporters, activists, lawyers, officials, and the people who protect them, the practical lesson is blunt. Treat advertising as executable third-party content. Block it on sensitive devices and at the network boundary where you can. That measure reduces one route. It does not clean an infected device, identify the operator, or make a high-risk person invisible.

## What investigators found in the ad stream

Citizen Lab calls the technique “AdFection.” Its [15 October report describes the first public technical confirmation of zero-click mercenary spyware delivered through targeted advertising](https://citizenlab.ca/research/how-discover-tech-turned-targeted-ads-into-a-pipeline-for-spyware/). The researchers say an Israeli company called Discover-Tech bought the ads and operated the bridge between ordinary advertising systems and several surveillance businesses.

The confirmed victim was Al Jazeera. Network logs supplied by the broadcaster showed a repeated sequence on targeted devices: contact with Discover-Tech infrastructure as an ad loaded, contact with a server attributed to Candiru that checked the device, and then communication with command infrastructure used by the spyware. Al Jazeera also recovered a Windows program from one compromised device. Citizen Lab found distinctive behaviour matching a Candiru implant it had analysed in 2021.

That evidence matters because one network request proves very little by itself. An ad impression can show that a vendor reached a browser. It does not prove infection. Citizen Lab connected the ad delivery, the screening server, the later command traffic, and a recovered implant. The report attributes the spyware to Candiru with high confidence. It does not name the government or other customer that selected the targets.

[Al Jazeera says its internal team first noticed unusual patterns on journalists’ devices in 2024](https://www.aljazeera.com/news/2026/10/15/israeli-ad-tech-firm-helped-spy-on-al-jazeera-devices). The pattern continued into 2025, and the broadcaster asked Citizen Lab to investigate. The network condemned the operation as an attack on journalists and press freedom, while stopping short of identifying who ordered it.

The number 45 also needs care. Citizen Lab says at least 45 Al Jazeera devices were targeted. Public reporting sometimes slides from “targeted” to “infected” as if the words were interchangeable. The investigators confirmed spyware infections in the operation, but their public wording does not establish that all 45 devices completed every stage. Keep the claim at the level the evidence supports.

The wider ad system carried much more traffic. Citizen Lab analysed nearly four billion Discover-Tech bidding records from a four-month period in 2026. Those records placed the company’s ads on almost 200,000 distinct websites and mobile apps across dozens of countries. The separate ad-security company Confiant [observed Discover-Tech ads in 228 countries and estimated roughly 10 million impressions per day](https://blog.confiant.com/p/adint-weaponized-ad-tech-for-zero-click-spyware-and-mass-identity-vacuuming).

Reach still does not equal compromise. An ad appearing inside an app or on a government network is evidence that the delivery channel got there. It does not prove that a named employee was selected, that exploit code arrived, or that spyware took hold. Confiant found the hand-offs it attributed to Candiru at far lower volume, from hundreds to the low thousands of impressions per day in its data. That pattern fits narrow selection from a very large pool.

Think of the advertising traffic as a crowded station. Millions pass through. The surveillance system watches the turnstiles, recognises one face, and opens a private door for that person. The crowd supplies cover and the selection system supplies precision.

## How a routine ad auction became a targeting system

An online ad usually reaches you through an auction completed before the page finishes drawing. The website or app announces an available slot. The announcement may carry an advertising identifier, rough location, device details, IP address, page context, and other attributes. Buyers submit automated bids. The winner sends content for your browser or app to render.

This system is called real-time bidding. The name makes it sound like a sale of screen space. It also distributes a stream of observations about devices and people. A company that sees enough bid requests can watch the same advertising identifier move between networks, cities, apps, and websites. The ad market needs recognition to target and measure adverts. A surveillance operator needs recognition to find a target again.

Citizen Lab says Discover-Tech operated a platform called Controlium. Its interface let customers build campaigns, specify users, choose countries, select ad-buying services, and attach a remote address that would be contacted when the ad rendered. The company appeared on “certified” or “approved” vendor lists maintained by Google and the advertising association IAB Europe from 2020 until early October 2026. Those listings helped it take part in the ordinary supply chain.

The ad itself could look harmless. When it loaded, embedded code contacted a Discover-Tech server and sent identifiers or network information. For most viewers, the server could return nothing dangerous. For a selected device, it could return more code or hand the browser to a system that inspected the device and prepared the next stage.

That selective response is one reason conventional ad scanning can miss the attack. A scanner in a data centre receives the harmless version. A reviewer receives the harmless version. The intended person, arriving from the expected address with the expected device, receives something else. Eskimi, one of the ad services named in Citizen Lab’s report, said the tags changed their response according to the viewer’s device, IP address, and location. Narrow targeting helped the campaigns pass automated checks.

Confiant found a more detailed version of the same trick. Its researchers first caught suspicious code in April 2026. The code gathered device and network characteristics, opened an encrypted channel, and waited for instructions. It could check browser capabilities, screen properties, network state, mouse movement, typing rhythm, touches, and page visibility. It could also receive and execute additional code inside the context where the ad ran.

A fingerprint has two jobs here. It helps the operator decide whether the device belongs to the intended person, and it helps choose an exploit likely to work against that browser and operating system. There is little value in exposing an expensive exploit to the wrong machine. Each unnecessary delivery gives researchers another chance to capture it.

Confiant also observed Discover-Tech code collecting first-party cookies on more than 1,000 publisher domains in contexts where the browser made those cookies available. The captured values sometimes included live session identifiers, email addresses, or names stored inside tokens. The researchers assessed that this collection could help maintain an identity map after a person deleted ordinary tracking cookies.

The final break-in still requires a vulnerability. Advertising data can find and profile the device. Script delivered with an ad can inspect it and stage code. To escape the browser or app protections and install spyware, the operator normally needs one or more software flaws. Citizen Lab did not recover the browser exploit used in the Al Jazeera operation, so the public report cannot tell us which flaw opened that door or whether it still works on current software.

That missing exploit is an honest limit. The researchers recovered enough surrounding evidence to document the delivery route and identify the implant. They did not recover every link. Good security reporting keeps both facts in view.

## The click was never the security boundary

“Do not click strange ads” remains useful advice against many scams. It does little against the operation Citizen Lab documented. The browser processed the ad because the page or app told it to. The target did not need to approve the connection, open a file, or press a button.

This changes where the useful boundary sits. A person cannot inspect a real-time auction before it happens. A journalist cannot recognise a malicious square that looks like a game advert, especially when the dangerous response appears only to a selected device. The control has to act before or while the third-party content loads.

Ad blocking therefore becomes a security measure for high-risk work, rather than a preference about clutter. Citizen Lab recommends blocking advertising and tracking on devices and at an organisation’s gateway. It also recommends browsers with built-in anti-tracking features, ad blocking, and support for Global Privacy Control. For individuals, its advice includes a trusted blocking extension, a privacy-respecting browser, disabling ad tracking, resetting the mobile advertising identifier, and considering network-level blocking.

The US government reached a similar conclusion before this case. As Citizen Lab notes, the National Security Agency and the Cybersecurity and Infrastructure Security Agency had already recommended ad blocking for government environments because malicious advertising can deliver harmful code. The October 2026 evidence turns a long-discussed route for advanced spyware into a documented one.

Blocking at one layer is fragile. A browser extension protects only the browser where it runs. It may not see ads inside mobile apps. An encrypted name lookup or direct connection can bypass a simple domain filter. A network blocker cannot inspect every encrypted exchange, and blocking lists lag behind new infrastructure. Some browsers also limit what extensions can block.

Use layers that fail differently. Block known advertising and tracking requests at the network gateway. Use a browser that reduces tracking on the endpoint. Remove advertising IDs from managed phones where the platform allows it. Keep the browser, operating system, and apps current so that a delivered exploit has fewer old flaws to use. Separate sensitive work from casual apps whose advertising libraries you cannot audit.

There is a cost. Some publishers deny access when they detect a blocker. Some pages break. Mobile apps may replace ads with subscription prompts or refuse to run. Network rules create support work when a legitimate service shares infrastructure with tracking code. High-risk organisations should make that inconvenience an explicit policy choice instead of leaving each reporter to negotiate it during a live assignment.

The honest promise is smaller than “safe browsing.” Blocking ads removes opportunities for this delivery method. It does not stop a malicious attachment, a message exploit, a compromised website, physical access, account theft, or a source who carries an infected device into a meeting. One route closes. The adversary still has others.

## Who should change behaviour now

Most people were not selected for Candiru spyware. The confirmed operation involved a newsroom and a commercial tool associated with government customers. A reader with no plausible state-level adversary should install updates and block intrusive tracking because both are sensible, then get on with life. Constantly treating a slow page as evidence of espionage creates fear without protection.

The calculus changes for a journalist covering armed conflict, a dissident facing a capable state, a lawyer representing sensitive clients, a government official, an embassy worker, or an investigator holding material that somebody would pay heavily to obtain. The device may expose more than its owner. It can reveal a source’s name, a meeting location, an unpublished story, or a network of colleagues.

Newsrooms need an organisational response because adverts arrive through shared infrastructure. One reporter can install a browser extension. That does not cover native apps, managed phones, a research workstation, or the guest network used by a visiting source. The security team needs to decide where adverts and trackers are blocked, which exceptions are allowed, and how logs are preserved when a suspicious sequence appears.

Editors also need to understand the operational cost. A reporter should not have to disable a security control to open a source’s local newspaper ten minutes before filing. Provide a separate low-trust browsing device or remote browsing path for sites that will not work under the hardened policy. Keep reporting notes, source communications, password management, and privileged accounts off that device.

Small civil-society groups may lack a security team. Start with a narrower design. Put sensitive communications on a current, managed device with few apps. Use a browser that blocks third-party tracking and adverts by default. Keep entertainment, games, and casual browsing on another device if the threat justifies the cost. The split is imperfect, but it cuts the number of strangers allowed to run code beside confidential work.

Families and colleagues belong in the model too. A surveillance operator may choose the least protected device near the target. A partner’s phone can reveal travel. An assistant’s calendar can reveal the meeting. A source’s old laptop can expose the draft. Raising the cost around one famous person while ignoring everyone who schedules and talks to them leaves a cheap route open.

Do not push specialised controls onto every contact without explanation. A source may be unable to buy another phone, tolerate broken websites, or learn a new system safely. Ask what information would cause the greatest harm if exposed, then spend effort around that exchange. Threat models allocate scarce attention. They should not become loyalty tests.

## What to do on a high-risk device

The first response is prevention. Make the device a poor place for an advertising-based exploit to land. The sequence below fits a person or small organisation with a credible targeted-surveillance risk. Large fleets should implement the same ideas through managed policy and test the result.

1. **Update every component that renders outside content.** Install current operating-system, browser, WebView, and app updates. Restart when the update requires it, then check the version instead of assuming the download completed. The Al Jazeera report does not identify the exploit, so there is no single patch number to chase.

2. **Block adverts and trackers in the main browser.** Use a browser with strong blocking built in or a trusted extension from the browser’s official store. Keep the blocker updated. Test several work-critical sites before travel or a deadline, and document the smallest exceptions required.

3. **Add a network layer.** For an office or managed home network, block known advertising and tracking domains at the gateway or DNS layer. Apply the policy to work-issued phones and laptops, including guest paths used for sensitive work. Monitor failure reports so people do not quietly bypass the control.

4. **Turn off personalised advertising and remove the advertising identifier.** On managed devices, enforce the setting through mobile-device management where possible. On personal devices, use the current privacy controls supplied by Android or iOS. Resetting an identifier can break one long-running link, but other fingerprints may reconnect the device later.

5. **Reduce app inventory.** Delete apps you do not need from the sensitive device, especially free utilities and games supported by opaque advertising libraries. Fewer apps mean fewer embedded ad exchanges and fewer background components receiving outside content. Review permissions after the cleanup.

6. **Separate risky browsing from protected work.** Open sites that require adverts on a different device or in a disposable environment that cannot reach source notes, newsroom identity systems, or password stores. Do not sign the low-trust device into the accounts that hold the material you are protecting.

7. **Prepare a clean communication route.** Decide how to contact a security responder without using the suspect device. Store the route somewhere available offline. For civil-society cases, [Access Now’s Digital Security Helpline](https://www.accessnow.org/help/) offers specialist assistance within its mandate.

8. **Keep useful network records.** Organisations should retain DNS, firewall, proxy, and endpoint telemetry long enough to reconstruct a sequence after an alert. Store the records away from the endpoint they describe. Citizen Lab’s appendix contains domains tied to Discover-Tech, while Confiant publishes additional indicators. Treat those values as investigative leads, not a complete detector.

9. **Practise the failure plan.** Run a short exercise in which a reporter’s device begins contacting suspicious advertising and command infrastructure. Decide who isolates the device, who protects sources, who preserves logs, and who communicates with the owner. A plan first opened during an incident is still a draft.

Check the result. Visit a test page that normally carries advertising and confirm that the browser and network controls block the requests. Confirm that work apps still function. Verify that managed phones no longer expose an advertising identifier where the platform offers that control. Record exceptions with an owner and an expiry date.

Avoid turning the list into a shopping spree. A fashionable privacy browser installed beside twenty advertising-heavy apps changes little. A commercial VPN may offer useful ad blocking, but it also becomes a party that can observe connection metadata. Choose it for a stated reason, examine its limits, and keep the network rule if it is doing real work.

## If you suspect this already happened

A strange advert, a warm phone, or a battery drop does not prove spyware. Consumer devices behave badly for ordinary reasons. At the same time, a clean home screen and normal battery life do not rule out a well-built implant. Start from evidence and consequence.

If your organisation finds a sequence that resembles the published indicators, preserve the logs outside the device. Record times, time zones, network addresses, device identity, browser or app version, and the rule that raised the alert. Do not begin by visiting the suspicious domains from the target device to “see what happens.” That changes evidence and may tell the operator that defenders are looking.

Move sensitive communication to a separate, updated device through a route the suspected device has not managed or backed up. Tell the smallest useful group. A broad warning sent through the same phone can expose the response plan and reveal which contacts matter most.

Do not wipe the device in the first frightened minute. A reset may eventually be correct, but it can remove the implant and the traces needed to understand who else was exposed. A qualified responder may want network logs, a memory or disk capture, browser artefacts, mobile diagnostics, or the device in its current state. Let the collection plan drive the order.

Account credentials need care. If spyware reached the browser, saved sessions or typed passwords may be exposed. Change critical credentials from a separate clean device, beginning with email, identity providers, password managers, and accounts that can reset other accounts. Revoke old sessions and review recovery methods. Preserve records of unfamiliar access before deleting it where that can be done safely.

Map the human exposure window. Which source names, meetings, drafts, travel plans, and contact groups were present? Which people face real harm if the operator saw them? Warn those people through another channel with facts you can support. Say that a device may have been compromised and that the scope is under investigation. Do not guess at the customer behind the spyware.

A newsroom should involve the device owner throughout. Technical staff understand logs. The reporter understands why one calendar entry matters more than a thousand browser requests. Editors and safety staff understand upcoming movement and publication risks. An incident response that removes the phone but ignores the reporting network protects the asset list and misses the people.

Published indicators can help an investigation, but they expire. Operators rotate domains and infrastructure. A hit needs context, and no hit does not clear the device. The stronger pattern in this case was a timed chain of contacts from ad infrastructure to a screening system and then to spyware control infrastructure. Behaviour travels better than one domain name.

## The advertising market supplied both cover and precision

The most important part of this story is larger than Candiru. Advertising systems were designed to identify a device, evaluate it in a fraction of a second, and deliver code chosen for that moment. The same features that make an advert “relevant” make the delivery route useful to an operator who already has a name on a list.

Scale provides cover. Confiant says Discover-Tech served about 10 million impressions each day. Citizen Lab found almost four billion bidding records in four months. A tiny set of spyware hand-offs can disappear inside that volume. The harmless ad is what nearly everybody sees, including many automated scanners.

Precision protects the exploit. The operator can wait for a familiar identifier, address, device model, country, and browsing pattern before revealing the expensive part. Code can change its response when a security scanner arrives. The attack surface is shared across websites and apps, while the dangerous payload is shown only to the selected machine.

Industry approval did not supply a trustworthy boundary. Citizen Lab found Discover-Tech on Google and IAB Europe vendor lists for roughly six years. Google terminated the company’s access to its ad exchange in early October 2026 and deployed Safe Browsing warnings for known actor-controlled domains. IAB Europe removed it from its vendor list on 8 October. Beeswax removed it, Vungle paused access for review, and Eskimi blocked associated accounts and domains.

Those actions reduce current reach. They do not answer how a company with reported links to surveillance figures passed through the supply chain for years, or whether another bridge remains. Discover-Tech and Candiru did not respond to Citizen Lab within the report’s stated deadline. CBC also reported that it could not obtain a response from Discover-Tech, and said Google confirmed that it had blocked the company for systematic policy violations.

There is a governance lesson here. “Certified vendor” can mean that forms were completed and technical integration succeeded. High-risk organisations should not translate that badge into permission for third-party code to run beside protected work. Trust needs evidence, recurring review, and a boundary that still holds when the vendor lies.

Publishers face a hard bargain. Advertising pays for reporting, including reporting that exposes surveillance. Blocking every advert can remove revenue from the same newsrooms under attack. That economic conflict is real. It does not justify letting an unaudited chain of intermediaries execute code on a reporter’s sensitive workstation.

The answer for one newsroom may be to block ads on staff devices while serving them to readers. Another may move sensitive browsing through isolation. Regulators can remove personal data from bid requests, require meaningful vendor checks, and impose consequences when the supply chain carries spyware. Readers alone cannot repair the market by finding the correct privacy toggle.

## A smaller target, with honest limits

This case closes a gap between suspicion and proof. Researchers had warned that targeted advertising could locate people for surveillance and carry malicious code. In October 2026, Citizen Lab documented the route through a real spyware operation, and Confiant independently exposed the code and advertising footprint around it.

The finding does not mean everyone who saw a Discover-Tech ad was hacked. It does not identify the customer behind the Al Jazeera operation. It does not reveal the browser exploit. It does show that a selected device could move from an ordinary ad impression into a mercenary spyware pipeline without a click.

Respond at the level of your risk. Most readers should keep software current, reduce tracking, use reputable blocking, and avoid inventing an espionage case from every glitch. People with a credible state-level adversary should make advertising a blocked input on sensitive devices, separate casual browsing from protected work, retain useful logs, and arrange expert help before an incident.

For a newsroom, the boundary belongs at the fleet and gateway as well as the browser. For a source, it belongs around the smallest possible exchange of identifying information. For a family member near a target, it may mean keeping travel and contact details off an old, advertising-heavy phone.

You cannot inspect an auction that ends before the page appears. You can refuse some of its code, give it less identity to work with, and keep the device it reaches away from the material that matters most. That raises the price of watching you. It also leaves the honest limit in view: a determined operator can choose another door.

If this kind of plain-English security analysis helps, the newsletter sends one email per month. The signup is on this site.

## Sources

- [The Citizen Lab: Exposing AdFection: How Discover-Tech Turned Targeted Ads Into a Pipeline for Zero-Click Mercenary Spyware](https://citizenlab.ca/research/how-discover-tech-turned-targeted-ads-into-a-pipeline-for-spyware/), accessed 2026-10-17
- [Confiant: ADINT: Weaponized Ad Tech for Zero-Click Spyware and Mass Identity Vacuuming](https://blog.confiant.com/p/adint-weaponized-ad-tech-for-zero-click-spyware-and-mass-identity-vacuuming), accessed 2026-10-17
- [CBC News: Israeli firm used ads to enable espionage, Canadian researchers say](https://www.cbc.ca/news/world/israeli-firm-targeted-ads-espionage-9.7381922), accessed 2026-10-17
- [The Guardian: Online ads weaponised to deliver ‘zero-click’ spyware, researchers say](https://www.theguardian.com/technology/2026/oct/15/israeli-firm-digital-ads-enabling-installation-of-zero-click-spyware-researchers-say), accessed 2026-10-17
- [Al Jazeera: Israeli ad-tech firm helped place spyware on Al Jazeera devices](https://www.aljazeera.com/news/2026/10/15/israeli-ad-tech-firm-helped-spy-on-al-jazeera-devices), accessed 2026-10-17
- [Access Now: Digital Security Helpline](https://www.accessnow.org/help/), accessed 2026-10-17

---

## About the author

Kubilay Tunca — Senior Full Stack Developer and Author. Founded Cyber Security in Plain English to translate complex security concepts into clear, practical advice, and writes the accompanying books on security, privacy, secure development, and AI systems.

## Books by this author

- **The Digital Fortress** — Your Everyday Guide to a Safer Digital Life. A warm, plain-English guide for people with real lives and finite patience. Learn the handful of habits that genuinely protect your money, accounts, and family, and get honest permission to ignore the rest. [Amazon](https://buy.cyber-security-in-plain-english.com/digital-fortress) · [Details](https://cyber-security-in-plain-english.com/books/the-digital-fortress)
- **The Anonymity Playbook** — Digital Survival for Whistleblowers, Journalists, Activists, and Everyone Else. A practitioner’s field manual for journalists protecting sources, whistleblowers, and activists. It explains how the surveillance actually works, what each technique costs you, and exactly where it fails. [Amazon](https://buy.cyber-security-in-plain-english.com/anonymity-playbook) · [Details](https://cyber-security-in-plain-english.com/books/the-anonymity-playbook)
- **Secure Software Development** — Practical patterns for building secure software. A hands-on security guide for developers and IT professionals who ship real software. Build, deploy, and maintain secure systems without slowing down or drowning in theory. [Amazon](https://buy.cyber-security-in-plain-english.com/secure-software-development) · [Details](https://cyber-security-in-plain-english.com/books/secure-software-development)
- **The Secure Harness** — Shipping Production Code with AI Coding Agents. A calm, practical guide to letting agents do useful work inside boundaries you set, enforce, and audit. Ships with 15 copy-pasteable artifacts: hook scripts, permission configs, release gates, and MCP templates. [Amazon](https://buy.cyber-security-in-plain-english.com/secure-harness) · [Details](https://cyber-security-in-plain-english.com/books/the-secure-harness)
- **The AI Native Engineer** — Build, Evaluate, and Ship AI Systems That Work in Production. Sixteen hands-on chapters, one real product. Grow it from a single model call into a retrieved, tool-using, observable, production-grade system, with evaluation treated as a habit from the first feature. [Amazon](https://buy.cyber-security-in-plain-english.com/ai-native-engineer) · [Details](https://cyber-security-in-plain-english.com/books/the-ai-native-engineer)

Full catalogue with contents and intended audience: https://cyber-security-in-plain-english.com/books

_As an Amazon Associate I earn from qualifying purchases. Buying through these links costs you nothing extra and helps pay for the blog._