BADBOX 3.0: what to verify before connecting TV boxes at work
BADBOX 3.0 raises questions about connected device provenance and the trust placed in network traffic. Businesses can turn those concerns into verifiable purchasing, connection and monitoring criteria.
Lire en portugaisBy Pentesty
Turn the fraud report into a business decision
The November 21 BleepingComputer report revisits HUMAN Satori research published on November 18. Researchers describe inexpensive AOSP devices without Play Protect certification used for hidden ads and residential proxies without consent. The operation reportedly reached roughly 40 billion advertising bid requests a day at its observed peak. Some of the fraud described did not even require an infected device. Those figures are not a count of business victims.
For a company using displays in reception areas, shops or meeting rooms, the practical question is how to approve the device powering them. The research does not show that your organization’s network was compromised. Our proposal is to treat device admission, connection monitoring and purchased media quality as decisions with their own owners and evidence.
Set conditions for purchasing and connection
Before installation, record the manufacturer, model, supplier, system version and support period. Require a verifiable origin for the firmware and an update process maintained by the manufacturer. Check any claimed certification on the device and against the relevant documentation; a description in a sales listing is insufficient evidence of provenance.
Make an explicit decision about products without demonstrable maintenance: replace, return or keep them disconnected during assessment. Retain the approval and the identity of the unit examined. Checking Android updates helps organize that record, but a displayed patch date does not establish firmware integrity on its own.
Test isolation and record outbound connections
Place presentation devices on a network with permissions defined for their role. Use authorized tests to establish whether they can reach workstations, administrative interfaces or internal applications that should remain inaccessible. Record the source, destination, applied rule and outcome. A VLAN name or configuration screenshot does not replace evidence that the restriction works.
Associate DNS and connection records with the inventoried unit, preserving timestamps and context. Compare content playback, idle periods and a restart within a controlled observation window. Investigate connections incompatible with the approved function. A few minutes without alerts does not certify the device, and sharing a public IP address does not, by itself, identify who initiated an action.
Extend verification to media purchasing
For advertisers, the number of requests does not establish how much money a campaign lost. Ask the media partner for the assessed period, seller and application identifiers, invalid traffic classification and the relationship between billing and validated delivery. Assign responsibility for examining discrepancies and documenting any financial adjustments.
The assessment should combine signals and allow challenges to its conclusions. Rejecting all residential access can harm legitimate users; accepting traffic solely because it presents a familiar application is also a weak criterion. Keep advertising delivery integrity tests separate from corporate access checks, retaining records appropriate to each purpose.
Document approval or removal
If provenance or maintenance cannot be substantiated, preserve the necessary records and remove the unit from the network while arranging replacement. Do not treat a factory reset as sufficient evidence of cleanup. For an approved unit, document support, configuration, restriction tests and the owner of the next review.
This criterion complements our device trust analysis following the Bazuka case. The useful outcome is a decision another team can check: which device may operate, under what restrictions and with which signals capable of triggering reassessment.
Frequently asked questions
Is every Android TV affected?
The report describes specific devices and infrastructure. Android in a product’s marketing name does not establish involvement; check the identity and provenance of the actual unit.
Do the 40 billion requests mean ads people watched?
No. The published metric is advertising bid requests at the observed peak. It does not equate to human views, unique devices or confirmed financial losses.
Is moving the device to another VLAN enough?
Segmentation needs to produce verified restrictions. Test the connections that must fail and the necessary services that must keep working, then retain the results.
Does the absence of a known indicator justify approval?
It is one observation in the assessment. Approval should also consider system provenance, support, network permissions and behavior monitored in the intended use context.
Sources and references
- BleepingComputer: BADBOX 3.0 reporting, November 21, 2026
- HUMAN Satori: BADBOX 3.0 technical report, November 18, 2026
Original Pentesty analysis, published November 21, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.