BADBOX 3.0 ad fraud hits 40 billion daily bid requests

Cheap streaming boxes sitting in living rooms around the world were secretly running the largest ad fraud operation ever disclosed, generating about 40 billion fake bid requests a day at its peak. HUMAN Security says its Satori threat intelligence team has now disrupted the operation, known as BADBOX 3.0, which also rented out owners’ home internet connections without their knowledge.
At its height, BADBOX 3.0 accounted for more than 4% of all connected TV traffic observed by HUMAN’s defence platform, and its volume was more than three times that of the previous record holder, VASTFLUX. HUMAN said the peak was larger than the last five years of its published investigations combined. The findings were published on 18 November 2026.
Hidden ads on boxes compromised before first use
The operation ran on low-cost, uncertified consumer devices, mostly streaming boxes, that run the Android Open Source Project and were built in mainland China and shipped internationally. They are not genuine Android TV devices and are not Play Protect certified, HUMAN said — and in many cases the devices were compromised before their owners even plugged them in.
The infection arrived as a backdoor packaged as a system app. Once running, the box contacted a command-and-control server, received a list of fraud modules, and downloaded whichever ones the operators had chosen. The modules were kept in a catalogue hosted on public code repositories, with backup copies spread across other platforms and mirrors, so removing one copy did not stop delivery.
The ad fraud modules imitated video advertising software and played video ads in a hidden window, so nothing unusual appeared on screen. The resulting bid requests pretended to come from well-known, legitimate streaming apps, borrowing their identities to make the traffic look like ordinary household viewing. Much of it arrived from residential internet addresses, which helped it pass as genuine. HUMAN also found fabricated requests claiming a streaming service had inserted an ad into a viewer’s stream, when no viewer device existed and no ad had played.
The boxes were also rented out as proxies
The same catalogue could enrol an infected box in any of 11 residential proxy networks without the owner’s consent. Those services let paying customers route their internet traffic through someone else’s home connection, using the owner’s bandwidth and public IP address. Ten of the networks were commercially available. The eleventh, PXYEDGE, is assessed by HUMAN to be a rebrand of IpMoYu, the proxy network run by the operators behind the earlier BADBOX campaigns. Residential proxies are commonly used to disguise account takeovers, fake account creation and card fraud.
HUMAN linked the module catalogue to TXCZ, a Chinese electronics manufacturer, through several independent connections: the command server ran on infrastructure associated with the company, modules checked for TXCZ-specific device properties before running, and a company domain hosted payloads and supplied settings for the hidden ads.
A family that keeps coming back
BADBOX 3.0 is the third chapter of a family HUMAN has tracked since 2023. The original BADBOX, disclosed that year, involved more than 74,000 infected devices with a firmware backdoor based on Triada malware. BADBOX 2.0, in 2025, spanned more than 10 million devices across 222 countries and territories. The new report also found modules from vo1d, a TV-box botnet first disclosed in 2024, in the same catalogue.
A separate Google-led takedown of the XCPE botnet on 2 November also hit several domains that controlled parts of BADBOX 3.0, according to HUMAN. Google said XCPE had enrolled more than 1.3 million set-top boxes. HUMAN says it continues to monitor the operators for further adaptation.
How to check your own box
HUMAN recommends buying streaming boxes from established brands and using official app stores. Owners can check whether a box is certified by opening the Google Play Store and going to Profile, then Settings, then About. Google also maintains a public list of certified devices. A bid request is an invitation to bid on an ad placement, not a confirmed payment, and HUMAN has not published a dollar figure for the fraud — but at 40 billion requests a day, the scale of the deception is unprecedented.