Skip to content
NewCybercrime Report 2026 · Eight Chapters · A Post Is Not An Incident → Explore The ChaptersCybercrime Report 2026 → Explore
Security NewsNews Explainer

BADBOX 3.0: Hidden Ads and Residential Proxy Abuse

BADBOX 3.0 ad fraud hid video ads on infected TV boxes and used residential proxies. See HUMAN’s measured traffic, attack chain, and defensive checks.

By
Brandefense Research Team
Reviewed by
Brandefense Analyst Team
Published
21 November 2026
Reading time
9 min read
Connected TV box with a hidden ad stream and residential network path

BADBOX 3.0 ad fraud turned infected connected TV boxes into two revenue channels: unseen video ads and residential proxy exits. In research released on November 18, 2026, HUMAN Satori said the operation peaked at roughly 40 billion bid requests per day. At that peak, it represented more than 4% of connected TV traffic observed by HUMAN’s own platform, not 4% of global TV traffic.

The same research traces both channels to a maintained catalog of modules and shared infrastructure. For defenders, the useful question is how a device, an ad request, and a home IP can appear legitimate separately while serving the same fraud operation. This article follows that chain and separates measured traffic from claims about device counts or takedown impact.

BADBOX 3.0 Ad Fraud Starts With a Compromised Device

HUMAN describes low-cost, uncertified devices running the Android Open Source Project, rather than certified Android TV OS devices. A backdoor packaged as a system application contacts a controller for a catalog of modules. The controller selects modules, and the device loads them in the background. This design lets operators change what a device does after the owner has started using it.

The catalog includes loaders, residential proxy modules, and ad fraud modules. Its records provide multiple delivery locations, so removing one copy of a module does not remove every route. HUMAN found versions of this catalog on public code hosting and mirrored delivery services. The same infrastructure served modules tied to earlier BADBOX investigations, which is part of the researchers’ lineage assessment.

Sequence diagramSequence diagram with 4 actors (Infected TV box, Module controller, Ad market, Proxy network) and 4 messages: 1, Infected TV box to Module controller: Checks in for selected modules; 2, Module controller to Infected TV box: Supplies ad and proxy modules; 3, Infected TV box to Ad market: Sends requests under borrowed app identities; 4, Infected TV box to Proxy network: Offers its household connection.Infected TV boxModulecontrollerAd marketProxy network1. Checks in forselected modules2. Supplies ad andproxy modules3. Sends requests under borrowed appidentities4. Offers its household connection
  1. 1Infected TV box Module controllerChecks in for selected modules
  2. 2Module controller Infected TV boxSupplies ad and proxy modules
  3. 3Infected TV box Ad marketSends requests under borrowed app identities
  4. 4Infected TV box Proxy networkOffers its household connection

This sequence is our simplified reading of HUMAN’s observed module flow. It does not mean every false ad request came from an infected box. HUMAN separately observed server-side ad insertion spoofing, where the claimed viewing device did not exist.

This is a cyber threat intelligence platform use case: evidence from devices, network behavior, and fraudulent ad inventory needs to be correlated before a decision. The botnet glossary gives the broader model for coordinated devices, while this case adds two monetization routes on one device.

How Can a Video Ad Play Where Nobody Sees It?

In HUMAN’s lab, malicious system apps loaded legitimate video advertising software and rendered ads in transparent or off-screen players. The owner could see an ordinary TV screen while ad activity ran outside the visible area. Researchers disabled the hiding behavior in one demonstration and then saw the ads that had been playing out of view. The lab result supports the mechanism; it is not a count of affected households.

The malware also borrowed the identity of legitimate connected TV apps. It contacted the ad servers used by those apps and produced requests that named the apps, even though those apps were not running on the infected device. A buyer could then see a familiar app name, a residential IP address, and a video ad opportunity. Each detail looked plausible while the underlying inventory was misrepresented.

HUMAN says the roughly 40 billion daily requests at peak came through several streams. Some came from malicious Android apps disguised as system components, and another stream used a legitimate channel app’s identity. The largest share carried spoofed or fabricated traffic attributes. Treat the peak as a count of bid requests associated with the operation, not unique ads seen, devices infected, or money lost.

Where Do Residential Proxies Enter the Chain?

The second module family enrolls devices into residential proxy networks without the owner’s knowledge or consent. HUMAN identified modules for ten commercially available services and an eleventh network it calls PXYEDGE. The device owner’s connection becomes an exit point that another customer can use. The owner may pay for the bandwidth without knowing why the device is generating extra traffic.

HUMAN assesses PXYEDGE as a continuation of IpMoYu, the network described in its BADBOX 2.0 research. Its basis includes module lineage, overlapping infrastructure, and task servers. This is a researcher attribution, not proof that every PXYEDGE node is an infected BADBOX 3.0 device. The reported proxy storefront’s node count and the researchers’ measured connection peak also describe different units.

The ad and proxy channels reinforce each other. Ads requested through household connections can resemble ordinary viewing, while a compromised TV box can also supply proxy capacity. HUMAN reported that, for one abused app identity, about 40% of requests overall and 80% of US requests used addresses flagged as residential proxy exits. Those percentages apply to that app identity in HUMAN’s observed traffic, not to all BADBOX 3.0 requests.

Residential routing alone does not prove abuse. A legitimate proxy user can also appear behind a household address. Investigators need behavior and context. Check the claimed app identity, device model, authorized app relationship, and whether unrelated apps share source addresses.

What Makes This Traffic Detectable?

HUMAN found the heaviest invalid traffic concentrated in a small group of inexpensive, uncertified device models. Those models sent requests almost exclusively under one family of app identities, unlike ordinary connected TV devices with varied app use. The company also observed a near-daily volume surge at the same hour. One abused identity almost always reported an internal test-build browser identifier that consumer TVs should not report.

Advertising teams can compare claimed app identity with the app’s authorized sellers and declared inventory paths. HUMAN says the apps sending these requests had no business relationship with the apps they named. Where sellers.json or app-ads.txt records exist, that mismatch supplies another review point. It should be combined with traffic behavior, because a valid app name alone is insufficient.

For security teams, separate device telemetry from advertising telemetry. A box making unexpected background requests or consuming bandwidth while idle merits investigation, but neither symptom identifies BADBOX 3.0 by itself. At the ad layer, compare device model, app identity, proxy-exit classification, timing, and repeated addresses across apps. Preserve the request and device context before blocking so a false positive can be checked.

What Do the Numbers Establish?

HUMAN calls BADBOX 3.0 the largest publicly disclosed ad fraud operation by bid-request volume. Its stated peak was roughly 40 billion requests per day, exceeding 4% of connected TV traffic seen by its platform at that time. These are platform observations and a peak comparison. They cannot be converted into a global infection count, a total advertising loss, or a current daily rate.

HUMAN also reported 367,023 requests to a mirror of one module repository during one month. It explicitly cautioned that these are fetch requests, not unique devices; researchers and scanners may have made some. A device could fetch more than once, and other delivery routes existed. The mirror figure cannot serve as a lower bound for the number of infected devices.

In Brandefense Blackhole’s BADBOX-related dark-chat count, the August 23 to November 21, 2026 window contains 15 posts. The unit is posts matching the topic in that period, not distinct operators, confirmed infections, or a separate incident count. We use this as a research-interest signal only; it does not verify HUMAN’s traffic measurement or establish that a new campaign occurred.

How Does XCPE Relate to BADBOX 3.0?

HUMAN says its researchers worked in parallel on the XCPE disruption with other organizations. The investigations were distinct but related, and the XCPE effort included infrastructure that operated portions of BADBOX 3.0. That overlap should not collapse the two operations into one device total. Our XCPE Android TV backdoor report covers the separate campaign and its device evidence.

HUMAN says it disrupted BADBOX 3.0 advertising activity and protects participating customers from associated invalid traffic. The report also describes redundant module delivery and a proxy network that survived earlier exposure under a new name. Taken together, these claims support a narrower conclusion: particular traffic and infrastructure were disrupted, while the conditions that enable reinfection and adaptation remained a concern.

The manufacturer link is also a researcher assessment. HUMAN cites shared communications patterns, a preinstalled privileged launcher, device-specific checks, and repository history. Those links matter for supply-chain review, but they do not prove that every device from that manufacturer is compromised. Inventory decisions should use model, firmware, certification, procurement channel, and observed behavior rather than a manufacturer name alone.

What Should Device Owners and Fraud Teams Do Next?

For household and enterprise device owners, HUMAN advises buying from familiar brands, checking Play Protect certification, and installing apps only from official marketplaces. If a low-cost TV box behaves strangely, retirement may be safer than trusting an app cleanup. A preinstalled or privileged component can be outside the control of an ordinary uninstall workflow. Record the device model and procurement path before replacing it.

BADBOX 3.0 defensive review
  • Inventory connected TV boxes and Android-based media devices, including model, firmware, certification status, owner, and network segment.
  • Review devices with unexplained idle bandwidth or background ad traffic, then preserve logs before isolation or retirement.
  • Compare CTV bid requests by app identity, device model, proxy-exit classification, timing, and shared source address.
  • Check claimed app inventory against authorized seller records where those records exist, and verify anomalies with the publisher.
  • Separate XCPE and BADBOX 3.0 evidence in incident records; document which source supports each attribution.

For ad buyers, the highest-value check is the combination of identity and behavior. A recognizable app name should not override a device population with improbable request concentration or repeated proxy exits. For fraud teams, review automated account activity by what it does, even when it arrives through residential IPs. These are our defensive applications of HUMAN’s findings, not a claim that every proxy request is malicious.

Revisit the decision as new observations arrive. Keep the measured peak, the scoped proxy percentages, and the separate XCPE evidence attached to the original source. That record helps teams update controls without turning a traffic anomaly into an unsupported device count or a blanket block on legitimate apps.

Frequently Asked Questions

Does the 40 billion figure mean 40 billion infected devices?

No. HUMAN reported bid requests per day at the operation’s peak. It did not equate requests with unique devices or households.

Is PXYEDGE the same as IpMoYu?

HUMAN assesses PXYEDGE as IpMoYu under a new name, based on shared infrastructure, module lineage, and task servers. That remains a researcher attribution.

What should an ad buyer check first?

Compare claimed app identity with authorized inventory and traffic behavior. Review device models, proxy-exit addresses, repeated source addresses, and unusual timing together.

Did the XCPE disruption eliminate BADBOX 3.0?

HUMAN describes XCPE as distinct but related and says the effort disrupted infrastructure operating portions of BADBOX 3.0. The report does not establish complete elimination.

Sources

  1. HUMAN Satori BADBOX 3.0 research, November 18, 2026
Share this articleLinkedInXEmail
Brandefense logo

Stay Ahead of Digital Threats.

Take control of your digital security with an exclusive demo of our AI-driven threat intelligence platform.

Open The Full Demo Form →