BADBOX 3.0: Cheap TV Boxes Rent Out Networks, What to Do
TL;DR · the short version
On 18 November 2026 HUMAN Security's Satori team said it had disrupted BADBOX 3.0, the largest ad fraud operation by bid volume ever publicly disclosed: roughly 40 billion bid requests a day at peak, more than 4% of the connected-TV traffic HUMAN observed and over three times the previous record holder, VASTFLUX. Low-cost, uncertified Android Open Source Project TV boxes built in mainland China carry backdoors that pull modules from a catalogue on public code repositories. The modules play video ads in windows the owner never sees, impersonating legitimate CTV apps, and enrol the boxes, without consent, in eleven residential proxy networks, one of them PXYEDGE, which HUMAN assesses to be the BADBOX actors' IpMoYu network renamed. HUMAN links the catalogue to Chinese manufacturer TXCZ. A bid request is an offer, not a paid ad; HUMAN gives no device count or dollar loss. For companies, the risk is a box on an office or guest network turning your IP address into someone else's proxy exit.
On 18 November 2026, HUMAN Security's Satori Threat Intelligence and Research Team said it had disrupted BADBOX 3.0, which it calls "the largest ad fraud operation by bid volume ever publicly disclosed". At its peak it generated roughly 40 billion bid requests a day, more than 4% of all connected-TV (CTV) traffic HUMAN's platform observed. It runs on cheap, uncertified Android Open Source Project (AOSP) TV boxes built in mainland China, which play video ads nobody sees and are enrolled, without their owners' knowledge or consent, in eleven residential proxy networks. HUMAN links the module catalogue to TXCZ, a Chinese electronics manufacturer.
Who this is for: security and network leaders whose offices, branches, meeting rooms, hotels, clinics, schools or retail sites have screens, plus fraud teams who use "residential IP" as a trust signal, and ad buyers spending on CTV.
Based on HUMAN's full technical report and press release (both read on 20 November 2026). PlayCISO is not affiliated with HUMAN, Google or any company named here. Command-and-control domains, IP addresses and repository names from HUMAN's indicators section are not repeated; take them from HUMAN's report.
Confirmed vs not yet known (20 Nov 2026)
- Reported by HUMAN: the ~40 billion daily bid requests at peak; more than 4% of CTV traffic it observed; more than three times the volume of VASTFLUX; infected AOSP boxes that are not Android TV OS and not Play Protect certified; hidden video ads impersonating legitimate CTV apps; enrolment in eleven residential proxy networks; the disruption.
- A bid request is not a paid ad. It is an ad slot offered for auction. HUMAN has not published how many were won, how many ads played, or what advertisers lost.
- Attribution, as HUMAN words it: the catalogue is linked to TXCZ "through several independent connections"; PXYEDGE is assessed to be IpMoYu, run by the BADBOX actors, under a new name. Nokia researchers reached similar conclusions, per HUMAN. TXCZ's response, if any, is not reported. No charges or arrests are reported.
- Not disclosed: how many devices are infected, which retail brands or models they are sold as, and which legitimate apps were impersonated (HUMAN says testing found nothing malicious in those apps).
- Related but separate: a Google-led disruption of an operation called XCPE, with HUMAN, Plume, TrendAI, Lumen and Synthient, found more than a million infected devices in over 200 countries and territories, and took down several command servers that operated parts of BADBOX 3.0.
How BADBOX 3.0 works
- Compromised before you plug it in. HUMAN's CISO Gavin Reid: "If you bought one, it was compromised before you plugged it in." A backdoor packaged as a system app checks in with a command server, receives a list of modules and loads the ones the operators select.
- A catalogue built to survive takedowns. Modules sit on public code repositories, with backups on a second code platform, an operator file server and, for some, a free open-source CDN that mirrors repositories even after the source is removed. Module class names imitate Android system packages. HUMAN says antivirus coverage of the newer modules "was effectively absent" when it analysed them.
- Half one: hidden ads. Modules load real video ad SDKs and play the ads in a transparent player positioned off screen. The bid requests name well-known CTV apps the box does not even run, so buyers see what looks like premium inventory from a household IP address. HUMAN also saw spoofed server-side ad insertion (SSAI) traffic, where no device exists at all, sold through the same supply paths.
- Half two: proxyjacking. Most modules in the catalogue enrol the box as an exit node in one of eleven proxy networks: ten commercially available services HUMAN names in its report, plus PXYEDGE. HUMAN says PXYEDGE advertises about 200,000 available nodes, and a monitoring server it found showed unique connections peaking at about 340,000.
- The halves feed each other. For one impersonated app, about 40% of requests overall and 80% of US requests came from addresses flagged as residential proxy exits. Fraudulent ad traffic routed through real homes looks like real viewing.
Timeline
| Date | Event | Source |
|---|---|---|
| 2023 | HUMAN discloses the original BADBOX: a firmware backdoor on uncertified AOSP devices, more than 74,000 infected devices observed | HUMAN (background section) |
| Late Nov 2023 | The older BADBOX 3.0 repository is created, within weeks of that disclosure | HUMAN |
| 2025 | BADBOX 2.0: more than 10 million devices across 222 countries and territories; disruption with the FBI, Google, Trend Micro, Shadowserver and others | HUMAN (background section) |
| Jul 2025 | Google announces legal action against BADBOX 2.0 botnet operators | HUMAN press release |
| Jan 2026 | The older repository gets its last update the same day the newer one is created, which HUMAN reads as a planned migration | HUMAN |
| Mar 2026 | The PXYEDGE domain is registered (15 March) and version 66 of the "myic" proxy module, tied to IpMoYu, is first seen | HUMAN |
| Mid-2026 | HUMAN sees a steep rise in CTV ad requests claiming one legitimate app and arriving from residential IP addresses | HUMAN |
| 18 Nov 2026 | HUMAN publishes the BADBOX 3.0 report and announces the disruption | HUMAN press release |
Find and block them on your network
Most coverage stops at "check your living-room box". The enterprise problem is the same box on a corporate, branch or guest network, where its proxy traffic leaves through your IP addresses. Practical steps, consistent with HUMAN's findings:
- Inventory streaming and signage devices. Pull DHCP leases, NAC records and switch-port data for anything identifying as an Android or TV device. Look in meeting rooms, reception screens, digital signage, staff rooms, retail displays, waiting rooms and hotel or clinic rooms. Unknown hardware vendors in the MAC prefix and generic hostnames are the first filter.
- Check certification on each one. HUMAN says BADBOX 3.0 boxes "are not Android TV OS devices and are not Play Protect certified". Google's advice is to make sure a device is certified and built with Android TV OS. A box that is neither has no place on a business network, whether or not it is infected today.
- Look at its traffic, not its screen. The ads are invisible. A TV box that keeps many simultaneous connections to unrelated destinations, carries upload volume a streamer should not, or talks to proxy-service and code-repository hosts is behaving like a proxy node. Flow logs, firewall session counts and DNS logs are enough to spot it.
- Put screens on their own network with default-deny egress. Allow only the streaming, signage and update services they need. That alone breaks proxy enrolment and module downloads, and it keeps an infected box away from everything else.
- Block residential-proxy SDK services at the gateway for device networks. HUMAN lists the ten commercial services the modules enrolled devices in; your secure web gateway or DNS filter may already have a category for proxy and bandwidth-sharing services.
- Replace, don't clean. The backdoor ships on the device. HUMAN's advice for consumers applies: retire a device that behaves strangely. Write certified hardware into procurement for anything that will sit on your network, including what facilities, marketing and franchisees buy.
- Stop trusting "residential IP" in fraud scoring. If you defend logins or checkouts, the infected boxes are exactly what make attack traffic look like a household. HUMAN says it judges behaviour "regardless of the network path it arrives on"; your rules should too.
What security leaders should do this week
- Ask for a count: how many TV boxes, sticks and signage players are on corporate, branch and guest networks, and how many are certified Android TV OS or another managed platform.
- Move uncertified ones off the network now, and segment the rest with an egress allowlist.
- Add a rule to procurement and facilities policy: no uncertified streaming hardware, no "free" boxes bundled with signage or hospitality contracts.
- Tell the fraud and identity teams that residential proxy traffic now has a large, cheap supply of real household addresses behind it.
- If you buy CTV advertising, ask your verification vendor whether BADBOX 3.0 traffic is filtered pre-bid and post-bid, and check that sellers appear in the apps' sellers.json and app-ads.txt files, a gap HUMAN highlights.
- Tell staff, briefly, that a cheap "fully loaded" TV box is a risk to their home network too. Point them to Google's Play Protect guidance.
Put it to work
- Egress Proxy Lab: run a secure web gateway policy and decide what a device network may reach.
- C2 Server Detection Lab: practise spotting the regular check-ins and odd connection patterns that give an implanted device away.
- Network breach response checklist: if you find a proxy node on your network, scope what else it could reach.
- BADBOX 2.0 explained: the previous chapter, and the consumer-side checks.
- Smart TVs and residential proxy SDKs: how bandwidth-sharing SDKs reached mainstream TVs, with consent this time.
FAQ
What is BADBOX 3.0?
BADBOX 3.0 is the name HUMAN Security's Satori team gave to the latest operation in the BADBOX family, disclosed on 18 November 2026. Cheap, uncertified Android Open Source Project (AOSP) TV boxes built in mainland China carry backdoors that fetch fraud modules from a catalogue hosted on public code repositories. The modules play video ads the owner cannot see and enrol the device, without consent, in residential proxy networks. HUMAN calls it the largest ad fraud operation by bid volume ever publicly disclosed.
Does 40 billion bid requests a day mean 40 billion ads were paid for?
No. A bid request is an ad slot offered for auction. HUMAN says BADBOX 3.0 generated roughly 40 billion of them a day at its peak, more than 4% of the connected-TV traffic its platform saw. HUMAN has not published how many were bought, how many ads actually played, or a dollar loss.
How many devices are infected with BADBOX 3.0?
HUMAN does not give a device count for BADBOX 3.0. It says one month of statistics from a single backup mirror of one repository showed 367,023 download requests, which are requests rather than unique devices, and that the real number of infections "can be presumed to be significantly higher". For comparison, HUMAN counted more than 74,000 infected devices in the original BADBOX (2023) and more than 10 million in BADBOX 2.0 (2025).
Who is behind BADBOX 3.0?
HUMAN links the module catalogue to TXCZ, a Chinese electronics manufacturer, "through several independent connections": command-and-control infrastructure, a TXCZ domain that serves payload download routes and hidden-ad configurations, modules that check for TXCZ-specific build properties, a repository account name that incorporates TXCZ, and a related threat actor. It also assesses that PXYEDGE, one of the proxy networks, is IpMoYu, the network run by the BADBOX actors, under a new name. Nokia researchers reached similar conclusions, HUMAN says. No arrests or charges are reported.
How do I know if my TV box is infected with BADBOX?
Check whether it is Play Protect certified and runs Android TV OS; Google's Play Protect help page, linked from HUMAN's press release, explains how. HUMAN says BADBOX 3.0 devices are AOSP boxes that are neither. The hidden ads are invisible, so you cannot see them; on a business network look for the box making large volumes of connections to many unrelated destinations. HUMAN's advice is to buy from familiar brands, install apps only from official stores, and "treat a device that behaves strangely as a device worth retiring".
Why should a company care about infected TV boxes?
Because a box plugged into an office, meeting-room, signage or guest network becomes a residential-style proxy exit on your IP address. Strangers' traffic, including the account takeover, fake sign-up and carding attacks HUMAN says residential proxies are used to disguise, leaves through your connection. And for anyone defending a login page, a residential IP is no longer evidence of a human customer.
Sources
- HUMAN Satori Threat Intelligence and Research Team: BADBOX 3.0 Hides the Largest Ad Fraud Operation Ever Disclosed Inside Infected TV Boxes (18 Nov 2026; Louisa Abel, Will Herbig, Lindsay Kaye, João Marques, Vikas Parthasarathy, João Santos, Adam Sell, Mikhail Venkov). Primary source, with indicators and MITRE ATT&CK mapping.
- HUMAN press release: HUMAN disrupts BADBOX 3.0 ad fraud operation (18 Nov 2026), with quotes from HUMAN's CISO and Google's Android security team
- HUMAN: BADBOX 3.0 and the economics of threat adaptation (18 Nov 2026)
- Earlier HUMAN reports: BADBOX 2.0 and the original BADBOX (background; figures in this article are taken from the BADBOX 3.0 report's own summary of them)
- Google Play Help: Play Protect and Android TV, both linked from HUMAN's press release for owners checking their devices
This article is educational guidance for security leaders, not a threat-intelligence report. Every figure is HUMAN's; check its report for indicators and updates before acting.