Comcast Turns Millions Of Routers Into Motion Sensors - The Same WiFi Signals Can ID You With 99.5% Accuracy

Dear Comcast customers: The internet gateway in your living room is now a motion sensor. It's free of charge, and the data can go to law enforcement without further notice. What's more, researchers have already shown that these same signals can identify exactly who is in the room.

On Tuesday, Comcast unveiled Xfinity Shield, a home security platform built entirely around the WiFi network. Its centerpiece, WiFi Motion, turns the leased gateway into a motion detector at no extra charge. According to Comcast, a typical customer has roughly 36 devices connected to their WiFi.

WiFi Motion works because a human body moving through a room disturbs the signal field between the gateway and stationary connected devices, like a printer or a game console. While the feature is opt-in and - according to Comcast's rollout materials - does not track phones or reach through WiFi extenders, those limitations are choices Comcast made in software, not limitations of the technology itself.

Comcast says WiFi Motion works "without recording video, capturing images or identifying individuals."

The company is not shy about identification as a product. The $15-a-month tier, Xfinity Shield Select, adds AI-powered cameras that identify people, pets, packages and vehicles. The free tier's restraint is a line drawn one product SKU away.

The company's own terms spell out where the data can go. According to the fine print that drew criticism when the feature first appeared last year, Comcast "may disclose information generated by your WiFi Motion to third parties without further notice" in connection with law enforcement investigations. Users pointed out at the time that the only complete opt-out is returning the Xfinity router.

The true capabilities of this technology were laid out by researchers at Germany's Karlsruhe Institute of Technology (KIT) in findings presented last November at the ACM Conference on Computer and Communications Security in Taipei. Using nothing but the traffic from an ordinary WiFi network, their system identified people with 99.5 percent accuracy in a study of 197 participants, regardless of viewing angle or how the person walked.

The method exploits beamforming feedback information (BFI) - housekeeping signals that every connected device sends back to the router so it can aim its transmissions efficiently. Beamforming has shipped with consumer WiFi hardware since WiFi 5. Because this feedback travels unencrypted, anyone within radio range can read it and, using a trained model, convert it into a crude radio image of whoever is present.

The person being identified does not need to be carrying a device at all; it is enough that other people's nearby phones and laptops are talking to the router. "This technology turns every router into a potential means for surveillance," KIT's Julian Todt warned. He described a scenario in which someone who regularly walks past a café running a WiFi network could be recognized there later - by authorities or companies - without ever knowing it happened.

One limitation provides a reality check: the system works only after it has been trained on a person's radio signature. The 99.5 percent accuracy figure applies to re-identifying enrolled participants, not identifying a stranger cold. The attack succeeds when it can say 'the person recorded here is the person recorded earlier'. So this limitation disappears once anyone with sufficient reach starts collecting signatures at scale.

The standards process has already moved past the warning stage. In 2025, the IEEE ratified 802.11bf, an amendment that deliberately builds sensing capabilities into WiFi - such as presence detection, fall detection, and breathing monitoring - and early implementations are already appearing in silicon. While the KIT team is urging for privacy safeguards in that new standard, the technique they demonstrated needs none of it: it runs entirely on beamforming features that have been deployed in consumer routers for over a decade.

The researchers tested mitigations. Cutting the frequency of beamforming reports barely dented accuracy even at heavily degraded sample rates, and encrypting the feedback would require rewriting the WiFi standard and breaking backward compatibility with billions of deployed devices.

Meanwhile, the visible surveillance layer is in open retreat in America. Flock Safety, the license plate reader network that became the flashpoint of this year's anti-surveillance backlash, spent last week announcing guardrails - cutting default data retention from 30 days to seven, and instituting automatic lockouts for abnormal searches. The company's own justification undercuts the gesture: Flock says more than 90 percent of searches are completed within a week anyway. This comes after a summer of reports of misuse, more than 50 agencies and jurisdictions walking away from contracts since January, and people in multiple states destroying the cameras outright. The ACLU called the changes a "thinly veiled PR attempt" built on "largely hollow security promises."

A camera is at least a visible object: it sits on a pole, neighbors notice it, city councils vote on it, and when enough people get angry, it comes down. Flock needed a decade and roughly 120,000 cameras to cover American roads. The WiFi build-out was finished years ago - one living room, café, and office at a time - and there is nothing on a pole to point at.