ICE Taps into Nationwide AI-Enabled Camera Network, Data Shows
ICE Taps into Nationwide AI-Enabled Camera Network, Data Shows
Flock's automatic license plate reader (ALPR) cameras are in more than 5,000 communities around the U.S. Local police are doing lookups in the nationwide system for ICE.
You're viewing a single thread.
It’d be a shame if anonymous types started working on poisoning all publicly accessible cameras with ai poison pills that brick whatever model you try to run on the footage
I read that glow in the dark material will trigger an ir motion sensor. So don't plant small flags coated with glow in the dark paint across from the cameras because it will cause them to take and send thousands of useless images and make them think their camera is broken.
Most of them will trigger from reflected IR, which is easy to do with some metallic mylar. Those emergency blankets cut into strips should work like a charm.
I work in security engineering, including massive video systems. With any commercial unit made in the last 5 years and any software past entry level consumer grade this is a non-issue. Especially if someone is using descriptive visual search when pulling up video vs just scanning through every motion event.
As far as I'm aware these are basically just trail cameras, they snap a photo on motion and send it over mobile data to be processed server side for ocr. They claim they can also identify make and model and anything different like bumper stickers. I wouldn't be surprised if their object recognition is just people in India. I also suspect that their OCR is, or at least was provided through 3rd party api calls.
You'd be surprised how much can be done at the edge with current cameras. I'm not sure exactly how these particular ones are set up, but the other major players in the space (Axis,Bosch,Panasonic) all have pretty surprising levels of local compute dedicated to "AI", and leveraging external VMS platforms can exponentially increase capabilities. It's pretty idiot proof once set up, as it's aimed at desk jockies that monitor and report with minimal systems training.
Again, not this specific system but this stuff is far from sci-fi anymore.
Or, they'll just develop downstream garbage filters and effectively ignore the little flags. Sure, some energy will be wasted, but it won't be occupying too many analyst brain cells.
Source: I have such a setup at home. My camera goes crazy detecting motion in the dark, CPU usage goes up. Main thing I notice? CPU temp rises from 50C to 55C. That's it.
These are also on solar/battery and use cell data.
Might give them trouble in pre-dawn hours, might not depending on the design. I doubt the municipalities and government agencies pay much of anything for data usage.
Fun fact, most places the department of transportation pays nothing for the electricity that runs street lights - electric company just gives it to them unmetered - in exchange for good and valuable consideration like right of way usage.
While they're at it, why not just hack the government to reverse last year's election, amirite?
I know most of us loved Mr Robot and watching dinozzo and abby double team a keyboard and Wolverine getting a blowy and all that fun stuff, but that really isn't how things work.
These aren't off the shelf pre-trained models. The model is a big part of the company's product and, increasingly, the cost of training is being partially offloaded to customers under the guise of "tune the model to your data".
And IF we have a Bones situation where someone has inscribed a virus onto human remains to destroy a one of a kind machine or whatever: That is what version control is for. "Hmm. The May 2025 model isn't working. Okay, switch back to April"
Also, these "models" are a lot closer to just running OCR on a feed and logging which traffic camera saw one of the flagged license plates.
You watch too much tv. All you need is to degrade the quality of the recorded video on any camera exposed to the public internet enough for ai to have divergent results due to how ambiguous the images captured are. There are thousands of hobby projects that let you browse actual feeds from such cameras and usually that means you can get hardware metadata and in most cases change how the video is recorded by patching the driver running on the already publicly accessed cameras. Why make an exaggerated strawman argument while at the same time pretending you know better than everyone else?
This article says it is local cameras installed by local police that are being used for ice by the local police department.
Claiming anonymous could do anything about it by poisoning AI models is absurd. Then you call out the skeptic for watching too much TV?
Besides, Anonymous hasn't done anything significant in 10 years They dos'ed Israel last year. Did it do anything? Was one less Palestinian killed?
Ahh yes the local police! An infamous bastion of web security, IT infrastructure, and thinking long term. Who could ever crack the default passwords on their IT setup? How could we ever hope to social engineer these above average intelligence elite local cops into plugging in a usb drive to their work computer. This is all definitely impossible, no local police branch has ever been a victim of ransomware so we know for sure it can’t be done and deserves all the cynicism and comparisons to Hollywood movies from the 80s.
Of course we also know that by “anonymous types” I meant that one specific group of people you have in mind who did that one thing 10 years ago and not just socially conscious programmers with basic knowledge of social engineering and web infrastructure. That would be a ridiculous thing to mean of course.
It's your claim that Anonymous would do anything when they won't and do it by AI poisoning that's absurd.
If your initial claim was, "It would be a shame if someone hacked their local police." it wouldn't have sounded like you just watched Mr. Robot.
Which is not an "ai poison pill"
That’s fair, you probably won’t brick their ai model but it can make it useless for that particular camera output.
Which.. is basically worthless because of just how many cameras there are out there.
A "fun" exercise a couple buddies and I did a few years (... decade?) back was to just use an afternoon of plugging python packages together and scraping county traffic cam feeds to track someone, with their consent, over a few days. And it was ridiculously easy to get their schedule down basically day one and even get a LOT of data on who they were seeing or where they went after parking just based on when and where the car "disappeared".
And that is just publicly available traffic cameras. Not the giant mess of speed and red light cameras and all the other crap we have in a modern surveillance state.
So even if people are climbing traffic poles and midlining over to the actual boxes to smash them? Those are even less of an issue than normal outages from rain on a windy day.
In 2003 a friend and I were brainstorming what the next big disruptive tech would be and how we might get investment to start a company based on it. My conclusion at the time: cheap digital cameras. 22 years ago they were already cheap and high resolution enough to kill the film camera industry, and they've only continued on through today with color night vision, etc.
He did finally get investment and start his own company: automating regulatory paperwork for small companies that would be swamped in it without help.
Meanwhile, networked cameras are approaching "smartdust" levels of ubiquity. It's like living around the time of Gutenberg and seeing the world relatively smothered in printed text leaflets, hundreds of times as many pages of text in less time and for lower cost than scribes. The changes have only just begun, and people aren't really aware of how fundamentally life has changed as a result.