TRENDING
An empty toll plaza entry with several lanes under a steel canopy, green arrow signals above the booths and a toll fare board beside the road
November 6, 2026
How to Spread Requests Across Servers in Python With the Power of Two Choices and Avoid the Herd Effect of Least-Loaded Routing
Cross-shaped arrow slit set in a recessed brick arch in a lichen-spotted castle wall at White Castle, Monmouthshire
November 6, 2026
Microsoft’s PDD Kit Turns Code Review Into Intent Review, but Verus Accepts a Login That Answers Every Attempt With a 500 Under the Contract the Post Prints
Several dried pufferfish with their spines raised hanging in a stack at a seaside market stall
November 6, 2026
OpenSSH Patches a Password Bypass on musl Systems Such as Alpine, Where a 257 to 1,024 Byte Password Logs In to Password-Disabled Accounts, and musl’s Maintainer Says the Bug Is Not in musl
Black silhouette of a wooden hourglass on a ledge against a grey cloudy sky with blurred bare branches
November 6, 2026
How to Stop Hung HTTP Calls in Python With Timeouts and Deadlines, and Catch the Slow-Drip Response That Slips Under requests, httpx, and urllib
Woodland path closed off with red and white barrier tape, traffic cones and a warning triangle
November 6, 2026
OpenAI’s “A Thousand Fewer Paths to Root” Turns $1.2 Million of API Credit Into 948 Candidate Patches, but Mainline Linux Shows 24 OpenAI-Linked Commits Since Mid-September and the Bounty Math Assumes Payouts the Rules Do Not Offer
06 Nov 2026
SXZ.io SXZ.io
  • Home
Search the Site
Popular Searches:
Technology Amazon AI
Recent Posts
A chrome ticket punch lying on a bright blue surface, the tool that makes a ticket good for one use only
How to Secure a WebSocket Endpoint in FastAPI and Stop Cross-Site Hijacking, Leaked Tokens, Floods, and Slow Readers
November 6, 2026
Yellow stenciled words Mind the gap painted on a station platform edge beside the railway tracks
OpenAI’s Astra Report Turns Silent Safety-Task Underperformance Into a Matched-Pair Gap, but the Ethical-Doubt Result’s Bar Runs From a 4-Point Gap to a 3.6-Point Reversal, and the Launch Card Quoted AISI as Seeing Very Few Refusals
November 6, 2026
Black and white photo of a speckled stoneware honey pot with a lid and a wooden dipper standing in its opening, against a dark background
OpenAI Releases GPT-6 Cyber, Rated Critical, to Approved Customers: the Post’s 99.93% Scope Score Is Measured Against the Older Cyber Model, the System Card Puts Its Base Model at 0% Honeypot Exploitation, and One Outside Benchmark Has It Solving Fewer Challenges
November 6, 2026
SXZ.io SXZ.io
  • Home

Categories

Articles 285 Posts
News 289 Posts
Learning Hub 259 Posts
Home/News/Google Bans Android TV Box Maker Videostrong Over a Backdoor on 1.3 Million Devices, but the Takedown Holds Domains and Leaves the Implant on Each Box
News

Google Bans Android TV Box Maker Videostrong Over a Backdoor on 1.3 Million Devices, but the Takedown Holds Domains and Leaves the Implant on Each Box

Google banned Videostrong Technology over the XCPE backdoor on more than 1.3 million Android set-top boxes. Registrars hold 125 of the 151 listed names, but the implant stays on each box and a YOKATV...

November 3, 2026 11 Min Read
33

Google Threat Intelligence Group said on November 2 that it and its partners “dismantled the XCPE backdoor which enrolled more than 1.3 million set-top boxes into a botnet.” It assesses “with high confidence” that Videostrong Technology Co., an Android device maker that The Verge places in Shenzhen, developed and ran the backdoor “for financial gain,” and it has banned the company, along with any future subsidiaries or affiliates, from building and selling certified Android devices.

Table Of Content

  • What Google says it did
  • How the backdoor works
  • The counts measure different things
  • What the takedown did to the infrastructure
  • The YOKATV configuration host
  • Where the reports disagree
  • What owners and network teams can do
  • What is still not public

The takedown cut off the operator, not the devices. TrendAI, one of the partners, says removing the two central servers “cut the operator off from the devices, although the implant remains on each box.” Google’s own description adds that the main component, as a system preload, “cannot be removed by the user.” On November 3 we checked every domain in Google’s indicator table against registration and DNS data. Registrars have put most of them on hold, ten names still resolve, and a configuration host that Plume ties to one line of the boxes still resolves to the address Plume gives for its broker.

What Google says it did

Google’s post lists four actions:

  • Legal action “to take down the domains and servers Videostrong used to control the botnet and install malware modules.”
  • Legal action against command-and-control domains “used by over a dozen malware families loaded as modules via this backdoor.” Those affiliates distribute their modules through other channels too, so Google assesses that the action “will sever access to at least 10 million additional compromised devices.”
  • A ban on Videostrong, and on any future subsidiaries or affiliates, from building and selling certified Android devices or supplying software to other certified vendors.
  • “Improvements to how we ensure the integrity of firmware on certified Android TV devices.”

The admission with the most consequence for other device makers concerns Google’s own certification. Google requires OEMs to pass “a comprehensive suite of compatibility and security tests,” but says “Videostrong employed obfuscation techniques to exploit a gap in our tests and load malicious code after the device’s first boot,” and that it has changed its review process. Its partners were Japan’s National Police Agency, Lumen, HUMAN Security, Plume, Synthient and TrendAI. Lumen says it is helping null-route the proxy servers involved and is publishing indicators in ThreatFox.

How the backdoor works

Google describes a three-stage design that Videostrong has run since at least late 2024. A stub preloaded in the firmware partition runs at the android.uid.system privilege level, under the name x.cpe or com.oemservice depending on the build. It fetches configuration over HTTP, then keeps an MQTT connection open to take orders: silent app installs, arbitrary shell commands, screen captures, and a setting change that disables Google Play Protect app verification. A loader, com.abc.nexus, runs the monetization modules. They come from other parties and turn the box into a residential proxy exit node, an ad-fraud engine or a further backdoor. “Since August 2026,” Google says, new dropper packages disguised as Android system apps have begun to replace that loader. Most infected devices are Android TV set-top boxes, with some telehealth devices, karaoke machines and digital signage.

The counts measure different things

The headline figure is “more than 1.3 million” in Google’s and TrendAI’s posts, with close variants elsewhere, but the figures underneath count different things.

Source What it reports What the number counts
Google “more than 1.3 million set-top boxes”; “at least 10 million additional compromised devices” Boxes enrolled in XCPE; devices reached by affiliates’ modules whose command-and-control domains were taken down (Google’s assessment)
Synthient “at least 1.3 million backdoored devices, with around half a million actively online” An estimate; the second figure is devices communicating daily
Plume 424,108 sessions on the main broker cluster and 26,477 connections on the YOKATV broker on September 8; unique devices “exceed 1.3 million” Concurrent connections read from the operators’ own statistics endpoint; the total is Plume’s estimate
TrendAI 1,472,554 devices across 206 models in “more than 170 countries” (June 3 to October 1); about 1.33 million scheduled for modules; 3,849,226 distinct exit addresses; 182,025 addresses available on a single day on average, 366,562 at the peak Lower bounds from intermittent telemetry
Lumen “over 1.3 million Android TV devices in more than 200 countries”; daily populations of Jaguar (nearly 2 million), SOAX (800,000), EarnFM (about 600,000), HailiangIP (over 500,000) and IpMoYu (over 200,000) Each proxy service’s whole botnet, of which Lumen says only “a large portion” comes from X.CPE

Read together, the 1.3 million is a count of devices seen over months, not of boxes online at once. Plume’s September 8 snapshot of the two brokers that had devices adds up to 450,585 concurrent connections (our sum), about 35 percent of the 1.3 million floor, which fits Synthient’s “around half a million actively online.” TrendAI’s average of 182,025 residential addresses available on a single day is about 12 percent of its 1,472,554 observed devices (our arithmetic). Lumen’s per-service populations cannot be added to the fleet total, because each describes a whole service, and Google notes that one device can run “as many as three residential proxy SDKs.” TrendAI’s top five countries are Malaysia, Brazil, Iraq, Mexico and Taiwan, with the United States, Germany, the United Kingdom, Australia and Canada each at roughly 26,000 to 42,000 devices.

What the takedown did to the infrastructure

We parsed the network-indicator table in Google’s post, which holds 145 domains, 6 subdomains, 11 GitHub account paths and 4 IP addresses, and skipped two wildcard rows. Starting at 10:29 UTC on November 3 we asked Cloudflare’s and Google’s public DNS-over-HTTPS resolvers for the address of each of the 151 domain and subdomain names, and the registry RDAP services listed in IANA’s bootstrap file for each registrable domain. No request went to any listed host.

Google’s label Names Under a domain on hold Still resolve
Honeygain / AppSalt 100 97 3
Moyu 17 8 1
Bird 7 7 0
Linkfog 6 3 0
XCPE and YokaTV hosts 5 4 0
Jaguar 4 2 1
Six other labels 12 4 5
All 151 125 10

The last row groups the Spirit, IPWeb, EarnFM, Fission, Gaganode and HiddenView labels.

All 123 held registrable domains carry the EPP status client hold, which RFC 5731 defines as “DNS delegation information MUST NOT be published for the object.” For 122 of them RDAP’s last-changed date falls between October 28 and November 2, and 113 of those on October 30. GoDaddy is the registrar of record for 108 of the 123. The two core domains, szbboys.com and the bootstrap domain anyevonline.com (which Google’s table covers with a wildcard row), are both held at Alibaba Cloud (Beijing), with registrations that run to February 24, 2027 and August 6, 2029. Under RFC 5731 a client status is one the sponsoring registrar can add or remove. The registry RDAP records show only the registrar, not the registrant or who asked for the hold, so we cannot say whether a court order stands behind these holds or how durable they are. Lumen describes emkmoh.com and io.pyuio.com as “seized by Google”; in RDAP their registrable domains carry the same kind of hold, at Cloudflare and GoDaddy, with last-changed dates of October 30, so what Lumen calls a seizure shows up in public data as a registrar hold.

Google says the operators hosted payloads as GitHub releases. GitHub no longer serves any of the 11 account paths in the table (HTTP 404 from its API at 10:29 UTC), though a 404 does not tell us whether an account was deleted or suspended.

Ten names still resolve. Three are Google-listed AppSalt domains registered through Tucows (dramapi.com, winklenark.com and wogglenift.com); one is earn.fm, which Lumen says Google “targeted” but which shows no hold; the rest are a HiddenView domain, two Fission subdomains, a Jaguar subdomain, a Spirit domain and a .my name we could not check through RDAP.

The YOKATV configuration host

Plume found three pairs of configuration server and MQTT broker. The first pair uses anyevonline.com and szbboys.com, both held. The second serves the YOKATV line of boxes: configuration at tms.yokatvonline.com:8443 and a broker at usamqtt.szbboys.com, address 38.145.197.30, where Plume counted 26,477 live connections on September 8. The broker’s name is now dead because szbboys.com is held. The configuration host is not. At 10:29 UTC on November 3, tms.yokatvonline.com resolved to 38.145.197.30 on both resolvers, and yokatvonline.com, which Google lists with the note “Fallback C2 for system packages,” has no hold. Google’s table lists 38.145.197.135, a neighbor in the same /24 block, but not 38.145.197.30. Plume says each configuration server returns its own MQTT host to the boxes. We did not connect to the address, so we cannot say whether it still answers, whether the YOKATV boxes would follow a new broker name, or whether Google’s legal action reaches it.

Past disruptions in this market did not last. Lumen says IpMoYu, a service that installed malware on X.CPE-backdoored devices, “began operating under multiple names” after the BadBox 2.0 operation seized its primary domain, TrendAI warns that “operators of this kind have rebuilt on new infrastructure before,” and Google’s post says that “when one component is disrupted, operators quickly pivot to new loaders or rival monetization SDKs.”

Plume’s analysis of the update client it found in Videostrong’s STV-2000 firmware shows what a lapse would cost. The client acts on an unsigned XML file, and for the generated update host under szbboys.com Plume says “DNS is the only security measure left.” The registrations run to 2027 and 2029, which is the window in which a hold would need to become a seizure or a sinkhole.

Where the reports disagree

  • Which boxes. Google’s post names no brand; “mecool” appears once, inside the ro.mecool.device system property the malware checks. Synthient attributes the MECOOL brand to Videostrong and published 182 model strings, with the caveat that “A model name alone is not sufficient to confirm compromise.” Plume says “more than a dozen different devices appeared to use the same MQTT broker, with MECOOL devices making up the majority,” and its device table also covers white-label models, some carrying other companies’ names. Online, mecool.com, which Plume cites for a MECOOL product photo, was still up at 10:29 UTC, with 115 of its 116 product variants marked unavailable (we cannot tell when that changed), and videostrong.com still carries “Google Certified Android TV Box” product labels.
  • Cleaning a box. Plume concludes that the botnet’s “core infrastructure is gone” and advises “a full factory reset” because modules may have dropped further backdoors. TrendAI says “Owners cannot remove the implant themselves.” Synthient advises that consumers “permanently dispose of these devices.” Google says x.cpe, as a system preload, “cannot be removed by the user,” and that after an uninstall it “automatically reinstalls from the /oem/ partition upon reboot.” A reset wipes user data, and none of the reports says it clears a firmware stub.
  • The 10 million. BleepingComputer says the additional devices include “retail set-top boxes, karaoke machines, and telehealth hardware” and that “Google says that not all of them are Android devices.” We could not find either statement in Google’s post, which lists karaoke machines, telehealth devices and digital signage among the infected “Android-based IoT devices” in the XCPE botnet itself.
  • Countries and counts. Lumen says “more than 200 countries”; TrendAI says “more than 170.” TrendAI notes that “Counts differ slightly between partners depending on when each snapshot was taken and what each could see,” and that every figure it gives is a lower bound.
  • Who is in the proxy market. Google says the AppSalt SDK is “owned by Honeygain, which is in turn owned by Oxylabs,” and its table labels 100 of its 151 domain and subdomain names Honeygain / AppSalt. Plume says AppSalt “appears to be built from the Honeygain SDK codebase, but operates as an independent network using its own servers and infrastructure.” Google says the integration “explicitly calls the optIn() method which sets a consent flag without displaying any user interface,” and Plume found two builds of the loader, one with a consent dialog and one without, with the second on the devices. Lumen says SOAX, “a service that claims legitimacy,” kept installing proxies on X.CPE-backdoored devices, while TrendAI says of the LinkFog and DianxinCloud services that they are “not necessarily malicious themselves” and that “The abuse lies in how the operator enrolled devices without their owners’ consent.” We found no response from Honeygain, Oxylabs, SOAX, EarnFM or Videostrong in the reports or coverage we read.

Each report also describes its own part of the operation and its own interests. Synthient’s was produced with the Motion Picture Association, whose executive is quoted on illicit streaming devices, and TrendAI’s closes by recommending its own products.

What owners and network teams can do

None of the reports gives a way to remove the firmware component from a box that is already in a home. Synthient recommends that the devices “be immediately pulled from online retail stores and marketplaces.” TrendAI advises consumers to avoid “unbranded or uncertified Android TV boxes and apps that pay for shared bandwidth.” Google tells users to keep built-in protections such as Google Play Protect active and to buy set-top boxes from reputable manufacturers, and points to its Android TV partner list. For companies, TrendAI says the exit addresses sat on consumer providers such as Comcast, Charter, AT&T, Virgin Media and Vodafone, “Blocklists of data-center ranges would not and did not catch this traffic,” and fraud teams “should stop relying on IP reputation alone.” It also says access “was sold as residential proxy and AI gateway services,” while adding that it observed no buyer and that nothing in its report is evidence of nation-state use.

To check a box you own, the reports give package names to look for: x.cpe and com.oemservice for the main component, com.abc.nexus for the loader, and, since August, droppers such as com.android.system.ntp and com.android.sso.auth.handler. With developer options and debugging over the network switched on (the menu names vary by box), a computer on the same network can list matching packages through adb. Its pm list packages command prints packages whose names contain the text you give it, as the Android Developers documentation describes. Run it once and switch debugging off again. An empty result for every name does not clear the box, because the operators have already renamed components once. Replace the address with your box’s.

adb connect 192.168.1.50:5555
for p in x.cpe com.oemservice com.abc.nexus \
         com.android.sso.auth.handler com.android.system.ntp; do
  echo "== $p"
  adb shell pm list packages "$p"
done
adb disconnect 192.168.1.50:5555
adb connect 192.168.1.50:5555
'x.cpe', 'com.oemservice', 'com.abc.nexus',
  'com.android.sso.auth.handler', 'com.android.system.ntp' |
  ForEach-Object { "== $_"; adb shell pm list packages $_ }
adb disconnect 192.168.1.50:5555

On the network side, the retry traffic should show up in resolver logs: the boxes keep asking for names that no longer resolve. Search your resolver or router log for the held names and match the client addresses to devices. We ran the loops above and the log search below against a stand-in adb and a sample log, not against an infected box.

grep -Ei '(^|[^a-z0-9-])(anyevonline|szbboys|yokatvonline)\.com' resolver.log
$names = 'anyevonline|szbboys|yokatvonline'
Select-String -Path resolver.log -Pattern "(^|[^a-z0-9-])($names)\.com"

What is still not public

  • The legal action. Google’s post names no court or case, and neither do the three outlets we read. A search of CourtListener’s RECAP archive for “Videostrong,” “anyevonline” and “szbboys” returned no results on November 3, though RECAP holds only filings that users uploaded.
  • Boxes already sold. Google’s stated firmware change is about “how we ensure the integrity of firmware on certified Android TV devices.” It does not say whether Play Protect or a firmware update will clear boxes already in homes, and Plume found a fleet that “mixes AOSP (Android Open Source Project) and Google-certified devices.”
  • Videostrong’s side. We found no statement from Videostrong or from the sellers of the affected boxes in the coverage we read.
  • The YOKATV configuration host. Whether it still answers, and whether it falls inside Google’s legal action, is not stated anywhere we looked. A similar completeness check on the Popcorn Time updater takedown found three of 20 listed domains outside the sinkhole.

Tags:

Android TVBotnetsGoogleIoT SecurityResidential ProxiesSupply Chain Security

Share

A yellow Atlantic canary perched on a leafy branch against a soft brown background
Previous Post

How to Load Untrusted YAML Safely in Python and Stop Code Execution, Alias Bombs, Duplicate Keys, and the Norway Problem

A navy blue circular floor marking in an airport hall showing a group of people gathered around a figure holding a flag, the meeting point symbol
Next Post

How to Build Rendezvous Hashing in Python and Keep Caches Warm When Nodes Come and Go

No Comment! Be the first one.

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Latest
06 Nov
How to Secure a WebSocket Endpoint in FastAPI and Stop Cross-Site Hijacking, Leaked Tokens, Floods, and Slow Readers
06 Nov
OpenAI’s Astra Report Turns Silent Safety-Task Underperformance Into a Matched-Pair Gap, but the Ethical-Doubt Result’s Bar Runs From a 4-Point Gap to a 3.6-Point Reversal, and the Launch Card Quoted AISI as Seeing Very Few Refusals
Trending
November 6, 2026
How to Secure a WebSocket Endpoint in FastAPI and Stop Cross-Site Hijacking, Leaked Tokens, Floods, and Slow Readers
November 6, 2026
OpenAI’s Astra Report Turns Silent Safety-Task Underperformance Into a Matched-Pair Gap, but the Ethical-Doubt Result’s Bar Runs From a 4-Point Gap to a 3.6-Point Reversal, and the Launch Card Quoted AISI as Seeing Very Few Refusals
November 6, 2026
OpenAI Releases GPT-6 Cyber, Rated Critical, to Approved Customers: the Post’s 99.93% Scope Score Is Measured Against the Older Cyber Model, the System Card Puts Its Base Model at 0% Honeypot Exploitation, and One Outside Benchmark Has It Solving Fewer Challenges
November 5, 2026
How to Spread Requests Across Servers in Python With the Power of Two Choices and Avoid the Herd Effect of Least-Loaded Routing
November 5, 2026
Microsoft’s PDD Kit Turns Code Review Into Intent Review, but Verus Accepts a Login That Answers Every Attempt With a 500 Under the Contract the Post Prints
November 5, 2026
OpenSSH Patches a Password Bypass on musl Systems Such as Alpine, Where a 257 to 1,024 Byte Password Logs In to Password-Disabled Accounts, and musl’s Maintainer Says the Bug Is Not in musl

Related Posts

Rows of server racks in a data center representing network infrastructure targeted by botnets
News

C0XMO Botnet Shows Why Old Router Firmware Still Matters

June 7, 2026
Close-up of a USB flash drive, representing physical data-theft risk in office security incidents
News

Fake IT Support Is Now Walking Through the Front Door

June 7, 2026
A phone security app on a smartphone resting on a laptop keyboard.
News

Everest Forms Pro Flaw Is Being Exploited to Create Rogue WordPress Admins

June 7, 2026
A phone secured by a padlock, illustrating AI data-leak containment and security controls.
News

OpenAI’s Lockdown Mode Is a Data-Leak Brake, Not a Prompt-Injection Cure

June 8, 2026
SXZ.io SXZ.io
  • [email protected]

Categories

Articles
Learning Hub
News

All Rights Reserved by SXZ.io ©2026