WIRED spoke with several experts about the explosion of surveillance technology, how police use it, and what the dangers might be. As tech advances, street cameras can now employ facial recognition and even connect to the internet. What does this mean for the future of privacy?
Transcript
[Narrrator] Amnesty International conducted a three borough census of surveillance cameras in New York and found more than 15,000 cameras in public spaces. This is just the part of the iceberg we can see. A 2021 survey found that about 17% of American respondents own smart security cameras. Many were private doorbell cameras.
[Albert] When you look at how many other brands of cameras there are out there, we’re probably talking about hundreds of thousands of cameras. These sensors have also become a lot more intelligent. They’re able to identify moving objects. All of which makes the data much, much richer. Your life can be rewound and your secrets can be revealed.
[Narrator] Wired spoke to several experts about the explosion of surveillance tech, how police use it, and what the dangers of it might be.
[Man] Immediate keyhole visual tasking. Target is on 20.
[Narrator] The surveillance ecosystem that could only be imagined back in the 1998 film Enemy of the State pretty much exists today, complete with smart cameras and high altitude aerial imaging. In the last two decades, cameras have gotten a lot cheaper, they’ve gotten smaller, they’ve gotten lighter. A lot of the NYPD camera systems have 4K resolution, night vision capabilities, 360 monitoring, or have a swivel mount. They could zoom in. Some of them are even positioned where they can see into the bedrooms of New Yorkers. We see growing drone deployment across the country.
[Arthur] The earliest predator cameras came in this giant sensor ball. That same sensing power is now available on drone cameras that weigh less than a kilogram.
[Narrator] Arthur Holland Michel’s book argues that domestic law enforcement’s adoption of militarized drones and wide area surveillance was born out of America’s wars in the Middle East.
[Arthur] War in the last two decades has actually come to resemble policing more than what traditionally was associated with conflict. The technologies themselves didn’t need a lot of reworking or recalibrating in order to have obvious use cases domestically.
[Announcer] A single red kite unit can continuously image an entire city size area in medium resolution.
[Narrator] This company’s marketing video shows an example of persistent Wide Area Motion Imagery, or WAMI.
It allows for the tracking of potentially hundreds of individuals simultaneously. So once an incident is observed, law enforcement can rewind and track persons of interest in the area all the way back to their homes.
[Arthur] If you want to go back weeks in the footage and see everywhere an individual has been, you can do that because the camera is always recording the entire frame.
[Albert] If you map the route you take to school, to work, to church, or to a mosque, you can’t get there unseen. And in the age of facial recognition, that means you can’t get there untracked. We know that the NYPD’s use of facial recognition is growing every year, and they’re not an outlier. That’s true for police departments around the country and it’s now increasingly common to see it used for graffiti, shoplifting, other low level offenses.
[Narrator] In fact, many major retailers have used or currently used image recognition cameras to monitor people in their stores. So, how does it work? Facial recognition works in very specific situations. For example, when we enter the airport in a passport control and it uses face rec to check our face against our passport ’cause it has to check one image against another image. We call this a one-to-one comparison. On a public camera, for example, if you search for a specific person, for example, a robber, you’re not doing a one-to-one comparison anymore but you do a one to possibly thousands of faces, and there it becomes less reliable.
[Albert] The problem is that facial recognition depends a lot on the quality of the photo. If you’re looking at someone straight on, if you have them well lit, if you have a high resolution image, it can be pretty accurate. But that’s not how most camera footage comes in from crime scenes. You get blurry images taken at night, low resolution from an off angle, just seeing the side of someone’s face.
[Narrator] Facial recognition is only as good as the data recorded and the database it’s being compared against. The first type of database used in facial recognition are the ones maintained by law enforcement.
[Arthur] There are also databases generated by Departments of Transportation, or DMVs, that collect people’s passport photo or driver’s license photo. There have been cases where the law enforcement agency has grainy CCTV image of an individual who they’re looking for and they may pass that to the DMV and run a facial recognition search.
[Narrator] And the third category of databases is potentially the largest of all: social media. We see facial recognition firms like Clearview AI going onto social media sites like Facebook, and Twitter, and Instagram and just scraping data, taking our images in bulk, downloading them, ingesting them into their database. Chances are that whoever you are, watching this right now, your image is in a Clearview AI database and there’s nothing you can do about it.
[Narrator] Police are increasingly turning to so-called fusion systems to help them connect the dots. Streamlining what once took countless hours of pavement pounding, video watching, and digging through various databases.
[Florian] And the more data there is out there, it becomes more and more complicated for our end users to really find information that they’re looking for.
[Narrator] Florian Matusek works for Genetec in Vienna. They make Citigraf, a fusion software system that pulls in various threads of data for law enforcement under a central display window. You could tell Citigraf, I want to see every time there was a theft reported in a radius of 500 meters. This gives you a very good filtering to find more relevant information, more relevant events by combining different data sources and show you this information on the map.
[Narrator] Another big player in the intelligence fusion space is Microsoft which helped expand the New York Police Department’s in-house machine learning fusion system, initially developed to prevent terrorism after 9/11.
[Arthur] The NYPD uses a product called the Domain Awareness System, effectively a fusion tool which consolidates all of the NYPD’s data that may be used in investigations in a single program in a single app that police officers can even have on their smartphones; footage from the city’s thousands of CCTV cameras, license plate readers, criminal record information. If a report of a crime comes through, they can open the software, pull up license plate reader to see what vehicles were nearby.
[Narrator] Not only is the data getting organized and the cameras getting smarter, they’re all connected to the internet now. Now that may seem like not such a big deal, but once a CCTV camera is connected to the internet, that information can be shared widely and instantaneously. That makes these enormous repositories of data available to really anybody who has the access code.
[Narrator] In their ongoing cataloging of surveillance cameras throughout New York city, Albert Fox Cahn’s group, STOP, discovered a potential security vulnerability. By searching the internet for the IP addresses of these cameras, what we found was one company, Hikvision, that had left its cameras unshielded. Most companies hide the location of their devices, but Hikvision didn’t. We found more than 16,000 Hikvision cameras in New York city. Hikvision is a Chinese manufacturer of internet enabled surveillance cameras. It’s controlling shares are owned by the Chinese government. That’s why some software manufacturers, like Genetec, avoid using Hikvision sensors for their systems. Certain Chinese camera manufacturers do not take type of security so seriously as they should. Any time that you digitize and connect surveillance data, you create an attack surface from actors both within and beyond.
[Narrator] So how do we watch the Watchers? How do we make sure that the images going from camera to cloud are protected? One line of defense is video encryption software. In real time, it pixelates all movement in the image. So you can imagine if there is an operator sitting in front of the screens watching the videos, the operator does not know who he or she’s watching. If really something happened and evidence is needed, then it’s possible to go back into the recording and based on user privileges to access the original video.
[Narrator] But even encrypted video can be unlocked by law enforcement, if they have a warrant. Sometimes the government doesn’t even need a warrant. We see police getting a direct link to these camera systems. So Amazon Ring’s partnership with police, thousands of agreements with the different police departments, setting up law enforcement portals where it’s easier for officers to identify the cameras in an area and to send requests for footage. It’s your choice. But the truth is people don’t feel free to refuse. And even if you’re not willing to hand it over, if it exists on Amazon servers or Google servers, they can hand over your footage, whether you like it or not. We all certainly want to reduce violent crime, especially when it appears to be on the upward trend, but we have to be very conscious of the costs of having these technologies. If you live in a city and you see that the mayor is using robotic systems or drones or artificial intelligence to watch the city, you are gonna think twice perhaps about going to a protest. You’re going to think twice about what you say or do.
[Albert] In the aftermath of George Floyd’s murder, we saw customs and border protection working with state and local police departments to deploy predator style drones to surveil protests.
[Narrator] Perhaps the most high profile case has been that of Derrick Ingram, who police accused of using a bullhorn too close to an officer’s ear. Police used a photo taken of him at a rally and identified him by matching it to social media.
[Albert] They stormed his block, sent heavily armed officers to surround his building, to intimidate him, to retaliate against him because of his activism all while holding a facial recognition printout in their hand. When we talk about this technology, it’s not about some abstract violation of our rights.
[Narrator] Further automation through so-called predictive policing systems have their own pitfalls. When you look underneath the algorithmic hood we see a lot of pseudoscience and Silicon Valley snake oil. And the more we collect and the more we combine, the more there is a danger that we, as humans, trust technology too much and think that it is perfect. So this is why we should never go in a direction where this information is combined automatically and decisions are being made automatically based on it. Part of what I find so dangerous here isn’t just that we have more cameras, isn’t just the way that they could be best used by police, but the fact that we have this growing culture of surveillance driven fear where people are constantly retreating behind their cameras worried more and more about their neighbors trained by these platforms to view them as a threat.
