BADBOX 3.0: Factory-Embedded Android Backdoor Leads to Hidden Ad Fraud and Residential Proxies
1. Basic Information Article Title: BADBOX 3.0 Hides the Largest Ad Fraud Operation Ever Disclosed Inside Infected TV Boxes Source: HUMAN Security Published Date: November 18, 2026 Collected Date: 2026-11-22T07:47
1. Basic Information
- Article Title: BADBOX 3.0 Hides the Largest Ad Fraud Operation Ever Disclosed Inside Infected TV Boxes
- Source: HUMAN Security
- Published Date: November 18, 2026
- Collected Date: 2026-11-22T07:47:22+09:00
- Original URL: HUMAN Security
- Related Source: BleepingComputer: BADBOX 3.0 Report
- Related CVEs: None
- Affected Products: Low-cost, uncertified AOSP TV boxes, phones, and tablets; devices associated with manufacturer TXCZ
- Response Priority: High (author's assessment)
- Reason for Response Priority: The author assigns high response priority because a system-level backdoor suspected of being factory-installed, combined with redundant payload delivery, turns devices into ad fraud and residential proxy exit nodes without user awareness. HUMAN observed traffic reaching approximately 40 billion bid requests per day.
2. Summary
HUMAN reported BADBOX 3.0, in which a system app backdoor on low-cost AOSP devices fetches modules from public repositories to play hidden video ads fraudulently and enroll devices into one of 11 residential proxy networks without authorization.
3. Attack Flow
From System App Backdoor to Ad Fraud and Proxy Enrollment
- A preinstalled/system app backdoor checks in with the C2.
- The C2 returns a module catalog and multiple download locations.
- The device fetches and loads modules from repositories, mirrors, CDNs, or operator hosts. Depending on the fetched module, the following related functions are executed (the list below does not indicate execution order):
- An ad module plays video ads off-screen and sends bid requests spoofing legitimate apps.
- A proxy module enrolls the device into a residential proxy network without authorization.
4. Attacker Position and Execution Location
- Operators capable of embedding system-level components into the device supply chain or factory images
- Remote operators managing C2, repositories, and payload hosting on the internet
5. Visibility for Victims and Administrators
Victims
- No anomalies are visible on the device screen, while ads play in a hidden window.
- Bandwidth and IP addresses are used by third parties as residential proxies.
Administrators
- Observable as system packages, high-volume fetches from repositories/CDNs, periodic CTV ad traffic, and proxy WebSocket connections.
6. Prerequisites and Failure Conditions
Prerequisites
- Presence of the backdoor/system app on a low-cost, uncertified AOSP device
- Reachability to the C2 and at least one payload mirror
- Dynamic module loading and the required system permissions are permitted
Failure / Mitigation Conditions
- Replace with Play Protect-certified, trusted devices
- Block IOCs and public repository/CDN delivery patterns at egress
- Isolate suspicious devices from the network and discard or re-image them rather than performing a factory reset (considering the possibility of firmware-level origin)
7. Impact Upon Success
- Ad fraud via hidden video ads
- Turning devices and residential IPs into proxy nodes
- Potential abuse of proxy exit nodes for account takeover, fake accounts, carding, etc.
- Consumption of bandwidth and power, and damage to network reputation
8. Observable Logs
- Email: Nothing notable
- Proxy / SWG / DNS: Connections to api.drgdj[.]top, file.drgdj[.]top, gh.drgdj[.]top, etc. / Repeated APK/JAR fetches from public code repositories and CDNs
- Endpoint / EDR: Module classes mimicking com.google.system.* / System apps such as com.android.systemsafe and com.shzhtxcz.txbox / Ad-related package com.speed.adv
- Identity / IdP: Nothing notable
- SaaS / Cloud: Nothing notable
- Network: Periodically spiking CTV ad bid traffic / Enrollment into commercial residential proxy services via WebSocket, etc. / Inconsistency where multiple app identities originate from the same IP
9. Determining Attack Success
- Confirmed malware execution or authentication success: Public information: HUMAN demonstrated hidden ad playback on test devices and observed related traffic in telemetry. They also analyzed the proxy module's enrollment function, though the backend rejected connections in the enrollment experiment published in the report. Approximately 40 billion bid requests/day is a platform observation value, not the number of infected devices. Scope: Devices and traffic determined by HUMAN to be related to BADBOX 3.0.
Separate attempts, execution, privilege acquisition, data access, and follow-on compromise; do not determine a successful compromise based on a single IOC or request alone.
10. Investigation Playbook
- Trigger: BADBOX 3.0 IOCs or anomalous ad/proxy traffic from uncertified AOSP devices
- Initial Checks: Check device model, firmware, Play Protect certification, system packages, and network segment / Preserve DNS/HTTP/TLS/flow logs and device images
- Endpoint: Check system apps, dynamic modules, launchers, persistence, and factory images
- Identity and Cloud: Check for account takeover or fraud alerts via the same residential IP
- Follow-on Activity: Track unauthorized access via proxy exit nodes, ad fraud, and additional module fetches
- Containment: Isolate devices, block egress, replace with certified devices / Hunt across the fleet for the same model/firmware
- Classification: Separate preinstalled backdoor, module delivery, hidden ads, proxy enrollment, and follow-on fraud
11. Defense and Detection Ideas
- Single Events: DNS/HTTP requests to BADBOX IOCs / Unknown system apps on uncertified AOSP devices
- Time-Series Correlation: Correlate C2 check-in -> module download -> hidden ad spike or proxy WebSocket
- Threat Hunting: Search for connections to drgdj/TXCZ domains, repositories/CDNs, and proxy services from the same model/firmware
- Logging Gaps: DNS, flow, and TLS visibility in consumer/IoT segments / Device inventory and Play Protect certification status
- Priority Controls: Eliminate uncertified devices, segment IoT networks, enforce egress allowlists, and monitor IOCs
12. Facts / Inference / Hypothesis
Facts
- Infected devices receive a list from the C2 containing module names, versions, classes/methods, parameters, and multiple download locations, and fetch payloads from public code repositories, other platforms, operator file hosts, and CDN mirrors.
- The ad fraud module borrows the identity of legitimate CTV apps and plays video ads in an off-screen transparent window to generate bid requests.
- In HUMAN's observations, BADBOX 3.0-related traffic peaked at approximately 40 billion bid requests/day, accounting for over 4% of CTV traffic on the Human Defense Platform. This is not an independent measurement of the number of infected devices.
- The proxy module enrolls devices into 10 commercial residential proxy services and PXYEDGE without the owner's consent. HUMAN assesses PXYEDGE to be a rebrand of IpMoYu.
- HUMAN linked the catalog to the Chinese electronics manufacturer TXCZ through multiple touchpoints, including C2, TXCZ domains, factory-installed launchers, device property checks, and repository accounts/history. This does not indicate Chinese government involvement.
Inference
- Due to payload mirrors, GitHub proxies, and multiple hosting locations, blocking a single domain or repository alone is likely insufficient to stop reinfection or module delivery.
Hypothesis
- None. Do not assert impact scope or attack success beyond published information.
13. MITRE ATT&CK Mappings
| ID | Technique | Confidence | Basis |
|---|---|---|---|
| T1071.001 | Application Layer Protocol: Web Protocols | high | Uses C2, repositories, CDNs, and HTTP(S) hosts for module delivery. |
| T1105 | Ingress Tool Transfer | high | Fetches modules from multiple download locations specified by the C2. |
| T1608.001 | Stage Capabilities: Upload Malware | high | Operators placed the payload catalog on public repositories and mirrors. |
| T1090.002 | Proxy: External Proxy | high | Enrolls devices as residential proxy nodes for commercial proxy services and PXYEDGE. |
14. Unknowns and Further Investigation
- Number of unique infected devices for BADBOX 3.0
- Scope of responsibility at each stage of the factory/supply chain
- C2, repositories, and proxy nodes remaining after disruption
- Number of affected devices in Japan
15. Impact on SOCs and Organizations
Organizations using uncertified AOSP TV boxes or digital signage devices should inspect Play Protect certification, system apps, outbound C2/high-volume ad traffic, and residential proxy service connections.
16. Summary by Audience
- SOC: Correlate drgdj-related C2, APK/JAR fetches from public repositories/CDNs, periodic ad traffic, and proxy enrollment.
- Administrators: Isolate, update, or discard uncertified AOSP devices, replace them with certified devices, and restrict egress.
- Users: Avoid low-cost uncertified Android devices, check Play Protect certification, and stop using devices exhibiting suspicious communications or behavior.
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes โ full credit and traffic to the original publisher.