As reported by BleepingComputer, the BADBOX botnet has resurged as BADBOX 3.0, generating an staggering 40 billion fraudulent ad bid requests per day — more than three times the previous record and accounting for over 4% of all connected TV traffic observed by HUMAN's Defense Platform. This follows a March 2025 takedown that severed 500,000+ devices from C2 servers and a Google lawsuit against the operators. The botnet is back, larger than ever.
Shield53 views this as a case study in why traditional botnet disruption strategies are structurally insufficient for IoT ecosystems. The pattern is now predictable: researchers identify a botnet, coordinate sinkholing or C2 severance, issue press releases, and within 12-18 months the operators rebuild with a new variant and a larger footprint. BADBOX has now followed this trajectory across three major iterations.
The Core Problem: Infected Hardware Remains in Circulation
The critical failure point is that takedowns target infrastructure — control servers, proxy networks, app store listings — but never the devices themselves. BADBOX infects cheap Android Open Source Project (AOSP) hardware: budget TV boxes, low-cost tablets, and generic streaming sticks sold through major marketplaces. These devices ship with malware pre-installed at the firmware level. Cutting them off from one C2 domain doesn't remove the malware; it simply triggers a fallback routine to contact a new server.
Until BADBOX 2.0, Google reported over 10 million infected devices. Even after the March 2025 disruption, those 10 million compromised units remained in consumer homes, in warehouses, and — critically — in retail supply chains. New shipments continued to introduce fresh infections. The botnet didn't need to recruit new victims; the supply chain did that automatically.
Residential Proxy Monetization Compounds Risk
BADBOX 3.0's residential proxy module deserves particular attention from defenders. HUMAN identified PXYEDGE as the successor to IpMoYu, with approximately 200,000 advertised proxy nodes and observed connections peaking near 340,000. This means compromised devices aren't just generating fake ad impressions — they're serving as exit nodes for potentially malicious traffic routed through residential IP space.
Organizations relying on residential IP reputation signals for access control, fraud detection, or geolocation enforcement should assume that a meaningful percentage of 'legitimate' residential traffic from affected regions may originate from BADBOX-controlled devices.
Who Is Affected
Shield53 Recommendations
- Inventory and isolate IoT hardware: Audit all Android-based TV boxes, streaming devices, and tablets on corporate networks. Any device from unverified manufacturers (T95, generic AOSP builds) should be quarantined pending firmware analysis.
- Block known BADBOX proxy ranges: Integrate HUMAN's published PXYEDGE and IpMoYu indicators into edge firewalls, WAFs, and fraud detection pipelines. Treat traffic from these residential proxy exit nodes as untrusted.
- Implement network-level ad-fraud detection: For organizations operating in ad tech or with significant streaming infrastructure, deploy SSAI request validation and bid request fingerprinting to identify synthetic traffic patterns.
- Procurement controls: Update hardware purchasing policies to require verified manufacturers, documented firmware provenance, and security certification for any Android-based device entering the environment. Budget Android hardware is the primary infection vector.
- Monitor outbound traffic anomalies: BADBOX devices generate sustained background traffic. Flag endpoints with persistent connections to unfamiliar ad networks or proxy infrastructure, especially during off-hours.
- Engage with takedown ecosystem differently: Industry coordination should prioritize device-level remediation — firmware updates, retailer recalls, and marketplace delisting — alongside C2 disruption. Sinkholing alone has now failed twice.
The BADBOX cycle will continue until the industry treats compromised IoT hardware as a supply chain problem rather than a network hygiene problem. Three iterations in, the evidence is clear: you cannot takedown your way out of a manufacturing-level compromise.