I worked on that.

It sends audio and/or video fingerprints (not frames, for privacy and bandwidth reasons), which are matched against a fingerprint database. Whatever people see on TV is usually 10 to 60 seconds behind the real live stream at the broadcaster (which is where the reference fingerprinting happens). GeoIP data can be used to roughly deduce where the TV is located, in order to better filter out false positives out of multiple matches (e.g. in the US where lots of programming on east/west side is just shifted by ~3 hours due to time difference).

are you saying that hypothetically, if the MPAA comes knocking on Roku's door with enough money and a fingerprint database of torrented movies/songs, Roku could then tell them they have people matching those fingerprints? After which I'm assuming they'd have enough justification to get a court order to get the contact info from Roku for matching users?

Possession, use, and downloading of copyrighted content is legal,* or no legal streaming services could exist. Copyright law is only concerned with licensing the transfer.

* in theory, as a civil matter, they could make you destroy any unlicensed copies, but they would have a hard time getting criminal charges pressed, as well as proving damages from watching a TV show from an unlicensed provider vs a licensed one

And yet I still can't fast forward/rewind when using Roku media player and dlna.

Is that mainly for ad attribution purposes?

Not for attribution. It’s for exposure. If Pepsi buys $1M in ads on NBC, it only knows the DMA and time slot/programs it bought the ads on. It doesn’t know the households it bought the ads on. With ACR data, it will know that you were exposed. From there, they can do a few different things. Audience studies (like they reached 2000 households with a certain income etc). Or they can run attribution studies. A company called Data+Math looks at exposure of these kinds of ads, understands which households weren’t exposed (as a control) and gives statistically significance calculations on linear TV ads to understand lift of sales (one example).

What's funny about this is that I think this is a legitimate and relatively non-evil use case.

It all comes down to lack of transparency/oversight and the option to exercise control as an individual.

Inscape, an ACR company, have this revealing paragraph on their blog. Note the "following your IP from the exposure to the ad, to the sales funnel" part:

"Advertisers like ACR data because it provides second-by-second feedback on how their ads are performing. Nielsen provides its data in 15-minute blocks, so if viewers tuned out after the first ad in a pod, the advertiser has no way of knowing. And since IP addresses are included, companies like iSpot.tv and Data + Math are able to use that information to create multi-touch attribution ratings that help advertisers understand how certain ads and placements helped move viewers through the sales funnel, from seeing the ad, to googling the product to actually buying it. It’s a lengthy process that requires a lot of data and a lot of rigor, but it’s an excellent way to prove to marketers that TV advertising actually works."

https://www.inscape.tv/resources/why-acr-data-is-poised-to-b...

They are spying on millions of people without their consent and without telling them about it. In what universe is this legitimate and non-evil?

I believe the parent isn't disagreeing with you.

Breaking down the parent's post:

""" What's funny about this is that I think this is a legitimate and relatively non-evil use case. """ - parent is saying that fingerprinting so the advertisers know who saw the ads is legitimate and relatively non-evil.

It all comes down to lack of transparency/oversight and the option to exercise control as an individual. """ - parent acknowledges that not telling the user and not making it configurable can be problematic.

"They are spying on millions of people"

If you consider tracking an anonymous identifier for the purposes of better marketing "spying" then I think that's a stretch. Calling out TV in particular for it is a bit silly - it's simply everywhere.

"...without their consent and without telling them about it."

Yes they are. You opt in or out when you buy the TV. They tell you about it then. You can be like most people and not read the fine print, but then don't be all surprised when someone's pulling the wool over your eyes.

> If you consider tracking an anonymous identifier for the purposes of better marketing "spying" then I think that's a stretch

If information about me or my machines is being collected without my express informed consent, that counts as spying.

Also "anonymous identifier" is a bit of an oxymoron. If the identifier is unique, then anonymity is not part of the equation.

I can't fathom the math and scale involved here making sense in the long term.

Eventually the marginal increase in profit is less than the marginal increase in adtech cost. I wouldn't be surprised if many industries passed that point years ago. There's probably a lot of hype and hubris disguising that fact, but someone's going to make a successful business case out of cheap, low-creepiness spray-and-pray advertising.

Depends on the manufacturer. Some use it to get you to use their other services. Some use it for second screen apps. Others for various on-screen info. Not sure of all use cases. I don't think the data was ever used for targeted tracking.

What is the video fingerprinting method used? Is it a publicly known algo? I was using a combination of "dhash" for individual frames and "simhash" to generate shingles for a bunch of videos and it worked "ok" but not as efficient as I wanted.

Did you ask user's permission for fingerprinting?

Any way to turn it off? Or perhaps block a specific domain via pihole?

Don’t let your TV access the internet at all.

Smart TV interfaces are almost uniformly worse than set top boxes (one or more of: bad UI, slow CPU, weird quirks, few updates) so you should avoid it anyway.

The current Apple TV (which I cite only because of familiarity) has a great UI, every major app, and robust HDMI-CEC support so you might never have to touch your TV’s remote again.

And Apple respects your privacy.

I have a TCL TV and use my Apple TV exclusively.

I was actually really pissed a while back because my in-laws were over and when I came home they told me "For some reason you hadn't connected your TV to the internet. We gave it your wifi password, and now it works!" Thanks. Now I have to change wifi passwords, and the power light on the TV constantly blinks because it thinks it should be connected to the internet, but isn't.

>And Apple respects your privacy.

And you really believe that?

Pretty much, yes. Because respecting my privacy fits their business model.

Consider that even the most trivial thing that makes Apple look bad gets leaked. If Apple was selling your private information, it would have leaked long before now. Also their financial reports show no indication of revenues that could be associated with private information marketing.

Nobody sells data, like pay and get hdd with data. They "analyze" it and sell results, or "allow access" for "optimization" of whatever. Or they have "partnership" and "exchange". Or they slightly obscure data (of course insufficiently) and then sell whatever resulted claiming that they don't sell "data". And so on, whole departments work full time on the ways to bullshit regulatory authorities into thinking that they don't sell personal data. (And they here I mean corporations in general).

And regarding Apple - I hear this "not their business model" argument often but I see zero real life reasons why it couldn't be but we wouldn't know it. It is like saying that "John only trades tomatoes, it is impossible to him to sell cucumbers, it is not his business model". How is even related, monster corporations have multiple divisions with multiple business models, one doesn't exclude another.

PS: this is for the sake of discussion. Personally I also tend to think that Apple collects much less data than FAGM, and there were experiments that indirectly support this theory. I'm thinking about moving to Apple ecosystem but it is rather costly and will cause vendorlock. Not an easy choice.

> Nobody sells data, like pay and get hdd with data. They "analyze" it and sell results, or "allow access" for "optimization" of whatever.

Yes, I think most people understand this and say "selling data" as shorthand (because, for a lot of people, it's a distinction without a difference).

Some smart TVs will join open networks if you don't give them one. And I expect that if 5G works as advertised you'll see surveillance capitalism adding 5G connectivity so you no longer have control over connectivity.

A website that catalogued the misbehaviours of the various smart TV operating systems (and the easiest methods of defeat) would be handy here.

E.g. Some TVs will honor wifi off setting. Or alternatively setting the TV to use the Ethernet port.

Or if it needs something on the other end, set up old underclocked Raspberry Pi as a basic router/DHCP server that connects to nothing; power it with TV's USB port.

If you've got a fancy router, connect it to your network with a fixed IP and firewall deny all packets from/to its IP.

If you've got a fancy AP, set up an alternative SSID that connects to an unused VLAN or otherwise routes to nowhere.

A Pi-Hole is good to have in every household. Takes minutes to set up and makes sure that queries to unwanted domains end up in the land of /dev/null

A pi-hole only works against adversaries that rely upon DNS, or haven't been coded to connect directly to "trusted" public DNS servers.

(I'm almost astonished that advertising networks haven't switched to using raw IP addresses everywhere.)

You could probably very easily make a list of the "evil" IP addresses if that happened

You can combine approaches of course. My main in-home DNS, per the DHCP settings on the wi-fi, is a Pi-Hole. Secondary DNS is the pfSense firewall, so nothing's dead in the water if the Raspberry Pi falls over for some reason.

The firewall has the same DNS block-lists as the Pi-Hole, but also has subscription lists of IPs to avoid. Most of those are spammers or malware, but can include whatever other category of malfeasance you desire.

Now we need some DIY guide on how to extract the 5G radio from your TV and turn it into an access point for a laptop.

This will depend on the jurisdiction. In GDPR land neither of this will fly as you obviously don't have consent. I own an Aldi TV which hasn't set up for internet connection. When I first started it I was greeted with a consent form which I declined. I am pretty sure that the setting I did (no internet) is honored both for PR and GDPR reasons.

With 5G, you will have the same problem. And I'd be very reluctant to buy anything stationary which has 5G connectivity.

Unscrew the back of your TV, the internal Wi-Fi antennas may well be be easily identified and disconnected.

e.g.

https://emfaction.userecho.com/communities/1/topics/124-disa...

Fun fact: If the signal is reasonably strong, Wi-Fi will sometimes still work even with the antenna disconnected.

If there is an unwanted and wide-open AP within range of an antenna-less smart TV, you have an unusual problem with countless fun and creative solutions.

I guess it's much less likely if the exposed antenna connector is properly terminated.

Or maybe grounding it? I bet if you mash a little ball of tinfoil into the connector, any residual chance of it working will evaporate.

Who knows? This could make a nice little experiment.

   1. Leave the antenna connected
   2. Unplug the antenna, leave the connector unterminated.
   3. Terminate the connector with ball of tinfoil.
   4. Use a proper impedance matched termination.
   5. Terminate with a proper impedance as close to the wifi chip as possible
   5./b Also cut the antenna trace on the PCB as close to the chip as possible
   Measure signal strength in all scenarios.

In the past I would have agreed with you on the poor quality of smart TVs. My Roku TV shatters all those expectations however, its fantastic. Great UI, plenty fast, no quirks I have found, and updates regularly.

I specifically bought a smart TV with Roku instead of whatever software Samsung/Sony is doing for these reasons.

I guess you ignored all the parents where they said that Roku sells your data.

> And Apple respects your privacy.

Yeah, exactly for their users in China.

We just provided the technology for major TV manufacturers. Most TVs allow you to disable it, although the feature may be called something unintuitive such as "Live Plus".

Create a fake username, setup a proxy with logger and connect your TV to internet via proxy. After couple of days/weeks just analyze all traffic and block strange hosts via /etc/hosts or Pi-hole.

AFAICT Roku sends logs to two separate endpoints, so blocking those visa pihole can give you some protection, however, it's hard to tell if any data is being sent to raw IP addresses.

It is not enabled by default. For the first time when you use a TV input, it asks you whether you want to enable it. If you have enabled it, you can opt out from settings later on.

It's probably using fingerprinting rather than uploading the whole frame.

And I would guess it's only audio fingerprinting, rather than full video.

This is a good reason for a pi-hole type of ad blocker on your network.

When I noticed my Roku TV was sending something to some remote analytics or tracking server every 30 seconds whenever it was turned on, I just blocked everything coming from it.

Eventually though I factory-reset it and didn't bother connecting it to the network at all. All the on-TV apps are junk and I'd rather just use an Apple TV (which sends it's own analytics, I know).

It is off by default and is enabled only if you opt in. They call it "More ways to watch". If you have enabled it, you can disable it later on from the settings.

The setting is controlled by "Settings/Privacy/Smart TV Experience". The first time you go into an input a dialog shows up asking the user to opt-in or out of this.

Are you sure about this? If the TV really is phoning home with whatever is on the screen, including anything personal the owner might happen to be displaying, that's a vast set of lawsuits waiting to happen. The GDPR fines alone could be staggering. I could easily imagine spyware logging whatever TV shows you're streaming and the like, but it's hard to imagine any business in this industry having lawyers dumb enough to allow sending actual screen images like that.

Sorry, I was being a bit lazy in my comment. I didn't specify, but I don't really suspect they are sending full frames back if for no other reason than bandwidth. But, honestly fingerprinting is so similar it might as well be the same thing. Though thankfully, yes, the fingerprint calculated for something personal probably is meaningless to them, but possibly could be replaced with a reversible option

One danger is that videos can now "phone home" with the TV they're viewed on. You could torrent through Tor and take all sorts of precautions, then watch on your TV at home and leak your viewing habits. Or worse, get someone else targeted for copyright enforcement if you watch pirated content on their TV.

How would data in the video be used to direct the TV to phone home? As in, what field would be set?

I guess it would be a watermark style change through all the video frames which affects the hashes - e.g. brightness or contrast or sharpness or some combination of that kind of thing - then seed that on torrent sites, and advertisers get to see which TVs watched the torrented film vs the official film.

I don't think fingerprints are the same at all. While still having privacy implications, fingerprints to match against broadcast content aren't uploading your family photo or caps from your home movie if that's what you're showing on screen.

It’s called ACR data and it’s very common. And yes, TVs are phoning home with it.

And GDPR only requires that you opt in. So when you sign into the TV for the first time, it gives you an opt in choice and many do it. The States is less regulated but will be soon.

GDPR can be quite strict on consent. See for example the UK's ICO guidelines. A sample of them:

> We don’t use pre-ticked boxes or any other type of default consent.

> We use clear, plain language that is easy to understand.

> We specify why we want the data and what we’re going to do with it.

> We give separate distinct (‘granular’) options to consent separately to different purposes and types of processing.

https://ico.org.uk/for-organisations/guide-to-data-protectio...

These are of course just guidelines, but if you don't explicitly inform your users that you will be sending images of what's on the screen over the Internet, you are likely to get in trouble. (And no, a giant EULA-type wall of text probably wouldn't be sufficient)

It isn't nearly as easy as you're suggesting to escape the scope of GDPR protections. There could be sensitive personal data or data about children involved. Even if it's just some identifiable individual in the screenshot, you still can't just rely on some sneaky "consent" as a blank cheque - that is merely a possible lawful basis for processing, and all the other provisions of the GDPR still apply.

Edit: Also, on your first point, ACR is generally a variation of fingerprinting technology. It wouldn't be sending entire screenshots of whatever is being displayed even if it's not broadcast content, at least not in any variation I've heard of. It was the idea of uploading the entire image that I was questioning before.

Fair point on the screenshot. Yea I dont think any TV is phoning home with screenshots. Rather it's phoning home with a processed signal of the A/V in order to ID that A/V. If there is A/V that cannot be recognized, it may phone home a screenshot of that.

And I also don't think it's easy to escape the scope of GDPR. I'm just saying companies come up with ways of being "GDPR compliant" and they've done so.

It may be sending video or audio frames but it’s also possible without it. The Chromecast can send side channel data to the TV.

Huh...I feel stupid. I should've realized that HDMI can send additional data... or rather should've put 2 and 2 together

I remember reading a comment some time ago which stated that they were concerned about the trustworthiness of HDMI, Thunderbolt, and the like as they can contain hidden features that could compromise the user. Fascinating. We need a firewall on our gateways, and on our peripheral ports.

That or it could be sending only a fingerprint (still a privacy issue, but less severe)

I'm really curious about this now, I wonder if there is some sort of way to sniff the the data / packets / whatever being sent and try to decode it