Flock Cameras Are Not Even the Worst Part of the System (Isaac Botkin)

Open linked video

> This outline was generated by an LLM from a transcript and may contain errors or inaccuracies. Prefer the original recording when available.

**Isaac Botkin and TREX Labs channel are not affiliated with this website. Their content is linked and outlined here because the message is worth sharing.**

The speaker argues that public debate over **Flock Safety** cameras is too narrow: the visible roadside cameras are described as only the collection layer for a much larger cloud database and analytics system. The central criticism is not just camera deployment, but the aggregation of vehicle, location, wireless-device, and inferred behavioral data into a unified surveillance product sold to law enforcement and potentially expandable to other data sources. The talk also argues that misuse risk is structural, citing broad access to data, weak audit practices, and searches allegedly conducted without case-specific justification.

Other notable threads include claims about camera hardware vulnerabilities, RF and Bluetooth collection, potential replacement of visible cameras with less visible sensors, comparison with **Axon** and broader surveillance-market competition, the use of “pattern of life” analytics, a cited query rate of **80–85%** without case IDs in some jurisdictions, and a London CCTV example used to argue that surveillance systems can be redirected toward monitoring ordinary people and protesters rather than solving serious crimes.

## New Business / Agenda Items 0:00

### Flock cameras as a public controversy 0:00

-   0:00 The speaker says **Flock Safety** cameras are receiving renewed public attention and that much public reaction is negative because of their potential misuse.
-   0:18 The main thesis is that the cameras themselves are only the visible “tip of the iceberg”; the more consequential issue is the company’s larger data and analytics system.

### Camera hardware and why the model was “inevitable” 0:43

-   0:43 The camera is described as a relatively simple remote device: camera, onboard computing, storage, battery, solar panel, LTE/cellular connectivity, and cloud upload.
-   1:24 The speaker argues that falling component costs made this type of solar-powered, wirelessly connected roadside camera commercially inevitable, and says **Flock Safety** is only the most prominent current vendor rather than the only one.

### Security and privacy criticisms of the devices 1:55

-   1:59 The talk claims the physical hardware has security weaknesses, including storage that is not secure if someone gains physical access.
-   2:14 The transmission path over cell networks is also described as insufficiently secure.
-   2:22 The speaker says the larger problem is the cloud database receiving and storing the camera data, which is presented as the real locus of misuse and the key distinction from ordinary standalone security cameras.

### Expansion beyond license plates into a multi-sensor surveillance system 3:29

-   3:29 The speaker says the devices have evolved beyond simple license-plate readers and now include video tracking and additional sensors.
-   3:40 The talk claims the cameras can detect Bluetooth and other wireless devices carried by passersby, allowing collection of phone and related device identifiers.
-   4:08 The company is described as moving toward drones and a “multi-sensor” fusion approach, with the strategic goal framed as building a unified database rather than merely selling cameras.

### Route mapping, inference, and “pattern of life” creation 4:32

-   4:32 The speaker describes how multiple cameras can be used to log traffic through different points and infer likely travel paths between them.
-   5:17 The criticism is that the database does not just store sightings at a time and place; it also creates inferred routes and movement histories.
-   5:36 The talk says the system also records or classifies vehicle characteristics such as color and make, broadening the searchable data beyond license plates.
-   5:48 The speaker points to reports of locating missing people not in vehicles as evidence that the system captures more than plate-only data.

### Potential face, gait, device, and vehicle telemetry linkage 5:59

-   5:59 The speaker warns of the potential to connect face recognition to license-plate and other data if video capability is sufficiently robust.
-   6:07 A video rate of up to **15 frames a second** is cited as enough, in the speaker’s view, to infer gait or skeletal movement patterns that can be tied to individuals or vehicles.
-   6:27 The talk says people’s phones, smartwatches, and earbuds emit unique wireless identifiers that could be captured and linked in the database.
-   6:57 The speaker adds that modern vehicles emit identifiers through tire pressure monitoring systems and says these signals are easy to collect.
-   7:30 The conclusion in this section is that the real product is the **Flock** database or operating system, with the cameras serving mainly as one collection method.

### Claim that visible cameras may be temporary 7:49

-   7:51 The speaker’s theory is that heavy criticism of visible roadside cameras may eventually push **Flock Safety** to replace them with less visible RF-detection or radar-based sensors.
-   8:03 The current units are described as temporary in practical terms because their batteries, exposed to heat and sunlight, may only last a couple of years.
-   8:22 The speaker notes cameras with a radar module and says its purpose is unclear because the module allegedly lacks an **FCC** ID that can be checked.
-   8:38 The concern raised is that **Flock Safety** could publicly concede concerns about cameras while continuing or expanding metadata collection through harder-to-detect sensors.
-   9:13 The speaker argues that if years of video and RF training data already exist, the company could claim camera images are no longer needed while still preserving movement and device tracking.

### Broader data-market ambitions and integration of other data sources 9:47

-   9:53 The talk says much of the same data ecosystem already exists through smartphones and modern connected cars.
-   10:18 The speaker says cars with **4G** or **LTE** modems already transmit telemetry, which manufacturers can sell onward, including to insurers.
-   10:31 The speaker’s theory is that **Flock Safety** is trying to build the largest law-enforcement database possible so it can later argue for access to additional mobile-carrier and vehicle-manufacturer data.
-   11:08 The key argument is that the business objective is long-term control of a large integrated data product, not merely roadside image capture.

### Competitors and surveillance-market consolidation 11:31

-   11:34 The speaker identifies **Axon** as another company positioned in this broader surveillance-data market through law-enforcement body cameras.
-   11:48 The point is that monopoly or near-monopoly control matters more in this market because the vendor with the biggest dataset and best training data may gain decisive advantage in analytics and predictive tools.

### “Pattern of life” as a policing product and why the speaker opposes it 12:17

-   12:17 The talk says “pattern of life” analytics are valuable because they connect people to devices, routes, and recurring habits in a searchable database.
-   12:38 The speaker argues this model favors clean, predictable data about ordinary, routine lives, while real criminal behavior may be more irregular and less suited to tidy profile building.
-   13:13 The criticism is that the system may become better at modeling ordinary people than identifying serious offenders, and that “outlier” behavior may be handled poorly or in ways that invite abuse.

### Misuse, access controls, and audit concerns 13:47

-   13:51 The speaker says a major recurring scandal around **Flock Safety** has been misuse of access to the shared data environment.
-   14:08 A key concern is that data from private locations, including **Home Depot** parking lots, can feed into the larger system and become accessible to law enforcement with permissions.
-   14:29 The speaker claims the company has been overly permissive in granting access rights.
-   14:35 The talk alleges many cases in which officers used system access for non-investigative purposes such as stalking or improper tracking.
-   15:01 The speaker further alleges that, in response, **Flock Safety** stopped saving the user ID associated with data requests, framing that move as reducing accountability rather than misuse.
-   15:17 The talk claims that in some jurisdictions about **80–85%** of data queries occur without a case ID or stated reason, which the speaker presents as evidence that much system use is not tied to specific criminal investigations.

### Critique of “crime-fighting tool” arguments 16:13

-   16:16 The speaker anticipates disagreement from both critics who think he is understating the danger and supporters who view the technology as an obvious public-safety tool.
-   16:50 The speaker rejects the claim that this kind of system materially helps officers solve concrete crimes for specific victims.
-   17:09 Instead, the talk argues that such systems naturally become tools for generalized surveillance and supervision, emphasizing collective monitoring over evidence tied to specific acts.

### Argument that metadata-only surveillance could be worse than cameras 17:42

-   17:45 The speaker says that if public criticism removes image capture but leaves metadata collection in place, privacy may not meaningfully improve and the resulting system may become more dangerous.
-   17:57 The argument is that camera images can at least provide potentially exculpatory facts, while metadata-based inference may place someone near a crime scene based only on estimated routes or device proximity.
-   18:26 The speaker warns that AI-driven graphing of relationships, routines, and associations could convert guesses and statistical probabilities into policing inputs.
-   19:02 The broader objection is to policing based on probabilistic inference rather than “hard facts.”

### London CCTV example and protest monitoring concern 19:20

-   19:20 The speaker uses London’s extensive CCTV network as an example of how surveillance systems can be misdirected.
-   19:32 London is described as having, for much of the speaker’s life, among the highest camera densities per capita.
-   19:53 The talk references published inquiries into rape-gang abuses and argues that the surveillance apparatus did not effectively stop or bring perpetrators to justice.
-   20:30 In contrast, the speaker says the system was effective at identifying people who attended protests against those abuses, illustrating what he sees as the more likely real-world use of mass surveillance.

### Broader criminal-justice critique and closing argument 20:47

-   20:47 The speaker says officers often want more tools, but argues that the central problem in many communities is not a shortage of surveillance technology.
-   21:02 The talk references **T-Rex** company policy against selling gear to law-enforcement officers in jurisdictions where private citizens cannot own the same gear.
-   21:19 The speaker argues that many officials do not need better pattern-of-life tools because the justice system already fails to act effectively against repeat offenders.
-   21:34 The claim is that serious crimes are often committed by people arrested **20, 30, 40** times and released, so this surveillance model does not address the root issue.
-   21:52 The closing conclusion is that the system is not a real solution to underlying community safety problems.

## Key figures and statistics

| Figure | Type | Context / topic | Timestamp |
| --- | --- | --- | --- |
| **LTE** | Network standard | Camera connectivity to cellular networks and cloud services | 0:58 |
| **15 frames a second** | Technical rate | Claimed video rate sufficient for gait or movement inference | 6:07 |
| **4G** | Network standard | Connected-car telemetry discussed as an additional data source | 10:18 |
| **LTE** | Network standard | Connected-car telemetry discussed as an additional data source | 10:22 |
| **80–85%** | Percentage | Claimed share of queries in some jurisdictions without case ID or stated reason | 15:32 |
| **20, 30, 40** | Arrest count | Claimed repeat-arrest pattern used to argue the justice system problem is not lack of surveillance tools | 21:41 |

## External vendors, brands, and software

| Name | Type | Context / topic | Timestamp |
| --- | --- | --- | --- |
| **Flock Safety** | Company | Central subject; roadside cameras and broader cloud surveillance database | 0:29 |
| **Home Depot** | Company | Example of private property using Flock cameras whose data can be shared into law-enforcement access | 14:08 |
| **Axon** | Company | Competitor or adjacent surveillance-market player through law-enforcement body cameras | 11:34 |
| **T-Rex** | Brand | Speaker references company policy on sales to law enforcement in restricted jurisdictions | 20:53 |

## Laws and policies

| Citation / name | Level | Type | Context / topic | Timestamp |
| --- | --- | --- | --- | --- |
| **FCC ID** | Federal | Regulation | Radar module discussed as lacking an FCC identifier that could be checked | 8:29 |
| **T-Rex policy on sales to law enforcement in jurisdictions where private citizens cannot own the gear** | County | Policy | Speaker cites company policy limiting law-enforcement sales in restrictive jurisdictions | 21:02 |

This outline was generated by an LLM from a transcript and may contain errors or inaccuracies. Prefer the original recording when available.

Isaac Botkin and TREX Labs channel are not affiliated with this website. Their content is linked and outlined here because the message is worth sharing.

The speaker argues that public debate over Flock Safety cameras is too narrow: the visible roadside cameras are described as only the collection layer for a much larger cloud database and analytics system. The central criticism is not just camera deployment, but the aggregation of vehicle, location, wireless-device, and inferred behavioral data into a unified surveillance product sold to law enforcement and potentially expandable to other data sources. The talk also argues that misuse risk is structural, citing broad access to data, weak audit practices, and searches allegedly conducted without case-specific justification.

Other notable threads include claims about camera hardware vulnerabilities, RF and Bluetooth collection, potential replacement of visible cameras with less visible sensors, comparison with Axon and broader surveillance-market competition, the use of “pattern of life” analytics, a cited query rate of 80–85% without case IDs in some jurisdictions, and a London CCTV example used to argue that surveillance systems can be redirected toward monitoring ordinary people and protesters rather than solving serious crimes.

New Business / Agenda Items 0:00

Flock cameras as a public controversy 0:00

0:00 The speaker says Flock Safety cameras are receiving renewed public attention and that much public reaction is negative because of their potential misuse.
0:18 The main thesis is that the cameras themselves are only the visible “tip of the iceberg”; the more consequential issue is the company’s larger data and analytics system.

Camera hardware and why the model was “inevitable” 0:43

0:43 The camera is described as a relatively simple remote device: camera, onboard computing, storage, battery, solar panel, LTE/cellular connectivity, and cloud upload.
1:24 The speaker argues that falling component costs made this type of solar-powered, wirelessly connected roadside camera commercially inevitable, and says Flock Safety is only the most prominent current vendor rather than the only one.

Security and privacy criticisms of the devices 1:55

1:59 The talk claims the physical hardware has security weaknesses, including storage that is not secure if someone gains physical access.
2:14 The transmission path over cell networks is also described as insufficiently secure.
2:22 The speaker says the larger problem is the cloud database receiving and storing the camera data, which is presented as the real locus of misuse and the key distinction from ordinary standalone security cameras.

Expansion beyond license plates into a multi-sensor surveillance system 3:29

3:29 The speaker says the devices have evolved beyond simple license-plate readers and now include video tracking and additional sensors.
3:40 The talk claims the cameras can detect Bluetooth and other wireless devices carried by passersby, allowing collection of phone and related device identifiers.
4:08 The company is described as moving toward drones and a “multi-sensor” fusion approach, with the strategic goal framed as building a unified database rather than merely selling cameras.

Route mapping, inference, and “pattern of life” creation 4:32

4:32 The speaker describes how multiple cameras can be used to log traffic through different points and infer likely travel paths between them.
5:17 The criticism is that the database does not just store sightings at a time and place; it also creates inferred routes and movement histories.
5:36 The talk says the system also records or classifies vehicle characteristics such as color and make, broadening the searchable data beyond license plates.
5:48 The speaker points to reports of locating missing people not in vehicles as evidence that the system captures more than plate-only data.

Potential face, gait, device, and vehicle telemetry linkage 5:59

5:59 The speaker warns of the potential to connect face recognition to license-plate and other data if video capability is sufficiently robust.
6:07 A video rate of up to 15 frames a second is cited as enough, in the speaker’s view, to infer gait or skeletal movement patterns that can be tied to individuals or vehicles.
6:27 The talk says people’s phones, smartwatches, and earbuds emit unique wireless identifiers that could be captured and linked in the database.
6:57 The speaker adds that modern vehicles emit identifiers through tire pressure monitoring systems and says these signals are easy to collect.
7:30 The conclusion in this section is that the real product is the Flock database or operating system, with the cameras serving mainly as one collection method.

Claim that visible cameras may be temporary 7:49

7:51 The speaker’s theory is that heavy criticism of visible roadside cameras may eventually push Flock Safety to replace them with less visible RF-detection or radar-based sensors.
8:03 The current units are described as temporary in practical terms because their batteries, exposed to heat and sunlight, may only last a couple of years.
8:22 The speaker notes cameras with a radar module and says its purpose is unclear because the module allegedly lacks an FCC ID that can be checked.
8:38 The concern raised is that Flock Safety could publicly concede concerns about cameras while continuing or expanding metadata collection through harder-to-detect sensors.
9:13 The speaker argues that if years of video and RF training data already exist, the company could claim camera images are no longer needed while still preserving movement and device tracking.

Broader data-market ambitions and integration of other data sources 9:47

9:53 The talk says much of the same data ecosystem already exists through smartphones and modern connected cars.
10:18 The speaker says cars with 4G or LTE modems already transmit telemetry, which manufacturers can sell onward, including to insurers.
10:31 The speaker’s theory is that Flock Safety is trying to build the largest law-enforcement database possible so it can later argue for access to additional mobile-carrier and vehicle-manufacturer data.
11:08 The key argument is that the business objective is long-term control of a large integrated data product, not merely roadside image capture.

Competitors and surveillance-market consolidation 11:31

11:34 The speaker identifies Axon as another company positioned in this broader surveillance-data market through law-enforcement body cameras.
11:48 The point is that monopoly or near-monopoly control matters more in this market because the vendor with the biggest dataset and best training data may gain decisive advantage in analytics and predictive tools.

“Pattern of life” as a policing product and why the speaker opposes it 12:17

12:17 The talk says “pattern of life” analytics are valuable because they connect people to devices, routes, and recurring habits in a searchable database.
12:38 The speaker argues this model favors clean, predictable data about ordinary, routine lives, while real criminal behavior may be more irregular and less suited to tidy profile building.
13:13 The criticism is that the system may become better at modeling ordinary people than identifying serious offenders, and that “outlier” behavior may be handled poorly or in ways that invite abuse.

Misuse, access controls, and audit concerns 13:47

13:51 The speaker says a major recurring scandal around Flock Safety has been misuse of access to the shared data environment.
14:08 A key concern is that data from private locations, including Home Depot parking lots, can feed into the larger system and become accessible to law enforcement with permissions.
14:29 The speaker claims the company has been overly permissive in granting access rights.
14:35 The talk alleges many cases in which officers used system access for non-investigative purposes such as stalking or improper tracking.
15:01 The speaker further alleges that, in response, Flock Safety stopped saving the user ID associated with data requests, framing that move as reducing accountability rather than misuse.
15:17 The talk claims that in some jurisdictions about 80–85% of data queries occur without a case ID or stated reason, which the speaker presents as evidence that much system use is not tied to specific criminal investigations.

Critique of “crime-fighting tool” arguments 16:13

16:16 The speaker anticipates disagreement from both critics who think he is understating the danger and supporters who view the technology as an obvious public-safety tool.
16:50 The speaker rejects the claim that this kind of system materially helps officers solve concrete crimes for specific victims.
17:09 Instead, the talk argues that such systems naturally become tools for generalized surveillance and supervision, emphasizing collective monitoring over evidence tied to specific acts.

Argument that metadata-only surveillance could be worse than cameras 17:42

17:45 The speaker says that if public criticism removes image capture but leaves metadata collection in place, privacy may not meaningfully improve and the resulting system may become more dangerous.
17:57 The argument is that camera images can at least provide potentially exculpatory facts, while metadata-based inference may place someone near a crime scene based only on estimated routes or device proximity.
18:26 The speaker warns that AI-driven graphing of relationships, routines, and associations could convert guesses and statistical probabilities into policing inputs.
19:02 The broader objection is to policing based on probabilistic inference rather than “hard facts.”

London CCTV example and protest monitoring concern 19:20

19:20 The speaker uses London’s extensive CCTV network as an example of how surveillance systems can be misdirected.
19:32 London is described as having, for much of the speaker’s life, among the highest camera densities per capita.
19:53 The talk references published inquiries into rape-gang abuses and argues that the surveillance apparatus did not effectively stop or bring perpetrators to justice.
20:30 In contrast, the speaker says the system was effective at identifying people who attended protests against those abuses, illustrating what he sees as the more likely real-world use of mass surveillance.

Broader criminal-justice critique and closing argument 20:47

20:47 The speaker says officers often want more tools, but argues that the central problem in many communities is not a shortage of surveillance technology.
21:02 The talk references T-Rex company policy against selling gear to law-enforcement officers in jurisdictions where private citizens cannot own the same gear.
21:19 The speaker argues that many officials do not need better pattern-of-life tools because the justice system already fails to act effectively against repeat offenders.
21:34 The claim is that serious crimes are often committed by people arrested 20, 30, 40 times and released, so this surveillance model does not address the root issue.
21:52 The closing conclusion is that the system is not a real solution to underlying community safety problems.

Key figures and statistics

Figure	Type	Context / topic	Timestamp
LTE	Network standard	Camera connectivity to cellular networks and cloud services	0:58
15 frames a second	Technical rate	Claimed video rate sufficient for gait or movement inference	6:07
4G	Network standard	Connected-car telemetry discussed as an additional data source	10:18
LTE	Network standard	Connected-car telemetry discussed as an additional data source	10:22
80–85%	Percentage	Claimed share of queries in some jurisdictions without case ID or stated reason	15:32
20, 30, 40	Arrest count	Claimed repeat-arrest pattern used to argue the justice system problem is not lack of surveillance tools	21:41

External vendors, brands, and software

Name	Type	Context / topic	Timestamp
Flock Safety	Company	Central subject; roadside cameras and broader cloud surveillance database	0:29
Home Depot	Company	Example of private property using Flock cameras whose data can be shared into law-enforcement access	14:08
Axon	Company	Competitor or adjacent surveillance-market player through law-enforcement body cameras	11:34
T-Rex	Brand	Speaker references company policy on sales to law enforcement in restricted jurisdictions	20:53

Laws and policies

Citation / name	Level	Type	Context / topic	Timestamp
FCC ID	Federal	Regulation	Radar module discussed as lacking an FCC identifier that could be checked	8:29
T-Rex policy on sales to law enforcement in jurisdictions where private citizens cannot own the gear	County	Policy	Speaker cites company policy limiting law-enforcement sales in restrictive jurisdictions	21:02

0:00 Now, I'm happy to see that flock cameras 0:02 are back in the news because they are 0:04 newsworthy. Uh, and I'm glad to see that 0:07 most of the opinions that I see being 0:09 published about them by the general 0:10 public are negative because the ways 0:13 that flock cameras can be misused are 0:16 pretty negative. However, uh even though 0:18 I don't want to take anything away from 0:20 the cautions that we have about the 0:21 cameras, the cameras are only the tip of 0:24 the iceberg. And so today I want to talk 0:27 about what the real threat is inside of 0:29 Flock Safety the company, but also 0:32 similar companies that are coming and 0:34 rather inevitably. 0:39 [music] 0:43 So this is a Flock camera and uh not 0:46 only is Flock Safety the company uh 0:48 relatively inevitable, this camera type 0:51 was inevitable. a remotely uh operated 0:55 camera that has a little LTE modem on 0:58 the side uh that lets it talk to cell 1:00 phone networks and communicate with the 1:02 cloud. Um they generally all at this 1:05 point have solar panels on them. Makes 1:07 them pretty easy to spot, but ultimately 1:09 the hardware of these different cameras 1:11 is not that different from your phone. 1:13 It's got a camera, it's got a brain, 1:15 it's got storage, it's got a battery. Uh 1:17 the solar panel is the only kind of new 1:19 thing and it talks to the internet and 1:21 puts its stuff in the cloud. As these 1:24 components got cheaper and more 1:26 ubiquitous, uh these solar powered 1:29 remote cameras with no connection uh to 1:32 the internet requiring infrastructure 1:34 and wires, just uh cell phone towers uh 1:37 off to the side somewhere uh were 1:40 inevitable. and cameras that wanted to 1:42 build these things and offer them as 1:43 services were inevitable as well. Uh 1:47 that's why Flock Safety, the company, is 1:50 the only person getting a lot of 1:52 attention for cameras like this right 1:53 now, but they're not the only player. 1:55 But the issue uh with these cameras are 1:59 many. Um the hardware of the camera 2:02 means that the storage device inside of 2:04 here that captures all the imagery is 2:06 not secure. Physical access to this 2:08 device is simple. uh and then the data 2:11 that is on there is accessible. Then uh 2:14 the way that that data is transmitted to 2:17 the cell phone networks uh also not 2:20 entirely secure. But the biggest issue 2:22 that I want us to focus on is the giant 2:26 database full of information up in the 2:29 cloud where all of the data captured by 2:31 the camera lives after it's transmitted 2:34 from the cell phone network. This right 2:37 here is where we want to spend a lot 2:38 more of our attention. Uh I would say a 2:40 lot more of our negative attention. This 2:42 is where the misuse happens. This is 2:44 what separates flock safety cameras from 2:47 regular uh security cameras. Uh this is 2:51 where we need to be attacking the misuse 2:54 of the data and the large database and 2:56 the AI models that manipulate the data 2:59 inside. If we get too focused on the 3:01 cameras, uh I think that we miss this. 3:04 Now, the cameras are easy to focus on 3:06 because they have m many problems. Uh, 3:09 and they're the physical manifestation 3:11 of flock safety in our communities. 3:14 They're mounted on this pole right here, 3:16 which is very fragile when you want to 3:18 cut it down and very not fragile when 3:21 you crash into it with your car. Um, 3:23 just the installation of these cameras 3:26 is a problem on top of everything else. 3:29 Um, and as these cameras have developed, 3:31 they have gotten new features. Um, they 3:34 have not just been taking pictures of 3:36 license plates, they can now track 3:38 video. They now have different sensors 3:40 on them, kind of like your phone, to 3:42 detect Bluetooth and wireless devices 3:44 like your phone going by. So now you 3:46 have the ability to collect all kinds of 3:48 data and put new kinds of data inside of 3:52 the database and then using fancy AI 3:54 models connect some of these things 3:57 together and make inferences about them 3:59 which we'll get to uh in a minute 4:02 because the RF data collection here is 4:04 just one of the many new tricks that 4:06 Flock is bringing to the table. They 4:08 have drones now. Uh the main goal is to 4:10 have a multi-ensor database uh and a 4:14 fusion system that ties all of those 4:16 things together. [gasps] So uh I guess 4:18 the question is why why is that a 4:22 problem? Well um there's a bunch of 4:25 different reasons. We'll get into some 4:27 of those in a second, but I want you to 4:29 think about all the data that can be 4:30 collected by this camera right now. 4:32 Originally the pitch was these were 4:34 license plate readers. And so as um we 4:38 have our database here and uh our 4:40 database grows to where it contains more 4:44 and more information uh it contains 4:47 information not just on license plates 4:49 captured but uh it begins to build out a 4:53 map of say different areas where traffic 4:56 is occurring. And by putting a flock 4:58 camera here and a flock camera here you 5:00 can grab all of the license plates all 5:02 the traffic that moves around. Once you 5:05 do that, you can now assume the paths 5:08 that cars were taken based on when their 5:12 license plates were detected on that 5:15 particular camera. And then that goes 5:17 into the database not just as a time and 5:19 a place. But you start to have database 5:22 entries that contain vehicle locations, 5:26 but also their roots. And some of that 5:30 has to be estimated because there's no 5:31 camera right here. You can see where 5:33 that starts to be an issue. But as the 5:34 cameras got more sophisticated, they 5:36 started to collect other things about 5:37 vehicles, not just license plates, but 5:39 color, vehicle make. These are all 5:40 things that Flock has advertised. So, we 5:43 know these things are relatively um 5:46 relatively understandable. Occasionally, 5:48 Flock will announce that uh they've used 5:50 these cameras to help find missing 5:52 people that are not in vehicles. And so, 5:54 we know that they're capturing other 5:56 stuff as well. the opportunity to 5:59 capture face recognition data, connect 6:01 it to license plate data, connect it to 6:03 other stuff, that's huge. If these 6:05 things have significant video capture 6:07 capability, let's say up to like 15 6:09 frames a second is enough to make uh 6:12 relatively good guesses about people's 6:14 skeletal structure and gate. Uh the way 6:16 that they move and walk around is a 6:18 thing that you can extract from raw 6:20 video now using AI models and you can 6:22 tie that to a person or to a license 6:25 plate on a car. Um, then of course the 6:27 person is carrying with them a device, 6:30 most likely probably a phone. Uh, maybe 6:32 they also have a smartwatch and maybe 6:35 they also have a earbuds. All of those 6:37 things uh emit signals that have unique 6:41 identifiers that you can connect to this 6:43 device. uh it will see those things 6:46 passing by, collect that data, send it 6:48 to the cloud and to this large database 6:52 and again fuse all of these different 6:54 things together and then the car itself 6:57 also has a number of other things. It's 6:59 not just the license plate, but uh your 7:02 vehicle almost certainly if it's uh 7:04 anywhere near modern has sensors inside 7:08 of each of the tires. Tire pressure 7:10 monitoring system. That's a wireless 7:11 system that emits unique IDs that can be 7:13 picked up really easily. Um I was doing 7:16 it a couple of weeks ago for reasons 7:18 that I will go into uh in future videos. 7:21 But all of this data can be collected by 7:23 the existing hardware very simply and 7:26 very easily. And all of that data can be 7:28 collected and put up here and fused 7:30 together. This database is the main 7:34 flock product. The flock operating 7:36 system is what is being sold, not just 7:39 the cameras. So yes, focus on the 7:41 cameras and all that they imply, but 7:44 don't lose sight of this larger tool 7:47 that is getting built. Uh the other 7:49 thing I want to point out is as there is 7:51 more and more negative attention on 7:53 these cameras, there's a greater 7:55 opportunity for Flock to use these 7:58 temporarily. My personal theory is that 8:00 these cameras only need to be temporary. 8:03 I mean, the ones that are installed now 8:04 are temporary. The battery that they 8:06 have inside of them out there in the hot 8:08 sun is only going to last for a couple 8:10 of years. And the more of these things 8:12 they can sell, uh the more cash Flock is 8:15 able to raise. But over time, they will 8:17 very likely replace these devices with 8:19 just RF detection. There are flaw 8:22 cameras out now that have a little radar 8:25 module on the bottom. We're not sure 8:27 exactly what it does cuz it doesn't have 8:29 an FCC ID on it that we can look up. Um, 8:32 but that is collecting RF data in some 8:36 fashion. And one of the things that 8:38 Flock could do as negative attention 8:40 grows, they could admit that they have 8:43 heard the general public and they have 8:46 gotten rid of the cameras. After all, 8:48 cameras do have weaknesses, not just 8:50 privacy weaknesses, but um they have to 8:53 see you, which means they have to be in 8:54 places where you can see them. And so 8:56 it's really easy to pick these things 8:58 out. But once you replace the camera bit 9:00 with just RF detection sensors and 9:02 radar, well then you can put these 9:04 things in more places and you can detect 9:07 more stuff in a way where the device 9:09 itself is much less detectable. And if 9:13 Flock has many, many years of video 9:16 training data alongside all of that RF 9:19 data and they've already built the 9:21 models to connect these things together, 9:23 they can and probably will make the 9:26 argument that the camera data is no 9:29 longer necessary, which is theoretically 9:31 a win for privacy, right? Because you're 9:32 no longer having your image collected. 9:35 But all of your data which is 9:37 theoretically all of your movements and 9:39 all of your activity that is still being 9:43 collected and there's more data that is 9:45 out there to collect. Obviously if you 9:47 have a sensor package that can collect 9:49 all this data then you can load it into 9:51 your database. Uh but even without these 9:53 devices you know a lot of this data is 9:55 being collected. your phone. Um, well, 9:59 it's uh it's it's a surveillance device 10:01 like like you know like the flot camera 10:04 and it is connected to these cell phone 10:07 devices all the time. Your earbuds and 10:10 your watch probably aren't unless you 10:11 have one of the newer watches, but if 10:13 you connect those to this device, all 10:15 that data can go into the cloud. Modern 10:18 cars have uh 4G or LTE modems that 10:22 transmit telemetry data up here and that 10:25 goes to mostly car manufacturers who 10:28 sell it to insurance companies. But uh 10:31 my theory is that the bigger this 10:33 database is that flock safety bills for 10:36 law enforcement purposes um the better 10:39 pitch that they have to the federal 10:41 government, other law enforcement 10:42 agencies, regulatory and legislative 10:44 groups that since they have the biggest 10:45 database, they should own some of this 10:48 other data as well. They should have all 10:50 of the mobile data that are collected by 10:53 the different carriers, all the location 10:55 data, they should have that in the big 10:57 database cuz the database is more useful 10:59 the bigger it gets. All of this car 11:01 information that is tracked by the 11:02 different manufacturers, well that 11:04 should live in the big database cuz the 11:06 bigger the database is, the better. So, 11:08 a lot of what Flock is doing right now 11:10 with their cameras, I think, is building 11:13 a tool that really has nothing to do 11:15 with cameras at the end of the day and 11:17 is a chance to grab market share and 11:19 make a bid for bigger data. And 11:22 ultimately, the data is going to be the 11:25 product, not the images. The images are 11:27 just the way that we collect the data at 11:29 the moment. Now, uh there are other 11:32 companies trying to get into this market 11:34 share. Um there is Axon. They make body 11:37 cameras now. They are the body camera 11:40 manufacturer uh in the United States. 11:43 Most of the body cameras used by law 11:44 enforcement are Axon cameras. Uh there's 11:48 some backstory there that's kind of 11:49 interesting, but grabbing some of this 11:50 market share so that you are the 11:52 monopoly makes sense when you're in this 11:55 space. There's a lot of areas where 11:56 you're just selling a product and any 11:58 market share is valuable. But when 12:00 you're going for being the sole provider 12:02 of the biggest database possible, having 12:05 the best AI models because you have the 12:07 best training data possible, uh you 12:09 really want to be the winner there. It's 12:11 uh there's not a lot of room for second 12:13 place, third place or fourth place. And 12:17 they talk a lot about pattern of life. 12:19 There's a huge value in developing a 12:21 database uh where every person is 12:24 connected to devices and movements and 12:26 they develop a pattern of life for this 12:28 person and for this person and for these 12:31 people over here. Uh and they pitch this 12:33 as a crimefighting tool. But there are 12:38 problems with this. If your goal is to 12:40 build a giant huge database that just 12:42 has really good, clean, orderly data 12:44 where patterns of life are clean and uh 12:47 predictable and your inferences are good 12:50 and solid because the data is nicely 12:52 joined and all of the fusion between 12:54 units uh works well then the outliers 12:58 become a bit of a problem and the 13:00 outliers are where a lot of your 13:01 criminal activity exists. Uh so this 13:04 pattern of life that gets developed and 13:06 these movements that can be adequately 13:08 tracked and well-guessed uh those work 13:10 well for people whose lives follow 13:13 patterns. Uh and a lot of the people um 13:15 that we want to do the you know crime 13:17 fighting activities against don't have 13:19 these same patterns. You know people who 13:22 steal a phone one day and trade it for 13:24 meth the next day in exchange for the 13:26 use of a car that's at a flop house and 13:28 they're only in it for maybe 12 hours 13:30 before something else is going on. these 13:32 guys don't have a pattern of life that 13:35 fits in this area. Now, an argument 13:37 could be made that looking for people 13:39 that don't fit into the pattern is good 13:42 police work, but I don't anticipate that 13:44 that is what is going to happen. And the 13:46 reason for that is the way that Flock 13:47 has handled some of their other 13:49 outliers. So, one of the big um 13:51 pervasive scandals that Flock Safety has 13:53 had to deal with is misuse of this data. 13:56 Because part of the issue is this data, 13:59 the big database is from many different 14:02 jurisdictions. And Flock puts cameras 14:05 and data access into private hands. Uh 14:08 Home Depot, for example, has lots of 14:10 flock cameras in their parking lots and 14:12 they have access to that data. They have 14:14 access to this subset of the data 14:17 because it's their property. However, 14:20 this Home Depot data is also accessible 14:23 to everybody in law enforcement with the 14:25 appropriate permissions. And Flock is 14:29 pretty liberal uh and lavish with the 14:32 permissions that they give. And so there 14:35 have been many cases, many many many 14:36 many cases where officers uh of the 14:40 peace are using flock access to do 14:44 things not related to crime fighting 14:46 activities. stalking of people, tracking 14:49 of people, unsavory sort of things. If 14:52 YouTube suggests you other Flock camera 14:54 content uh in your sidebar, you'll 14:56 probably find a bunch of these stories. 14:58 And the way that Flock has decided to 15:01 combat this is to no longer save the 15:04 user ID that requests data. That way, 15:07 people can no longer get caught doing 15:08 this sort of thing. No longer get caught 15:12 doing this sort of thing. Uh but even 15:15 with the information that we do have, we 15:17 do know that lookups of people um and 15:20 license plates and traffic and just 15:24 various pattern of life stuff happens 15:27 predominantly outside of specific case 15:29 files. So in some jurisdictions it's um 15:32 it's about 80 to 85% of all queries of 15:35 the data. Um there is no case ID, there 15:39 is no reason given for the search. the 15:41 vast majority of what's being collected 15:43 and what's being asked for does not 15:46 appear to be directly related to 15:48 catching specific criminals. Um, so 15:51 yeah, as as this develops and pattern of 15:54 life becomes a product that is sold to 15:56 law enforcement agencies or to 15:58 municipalities or to other types of 16:00 bureaucrats that are not actually the 16:01 ones working on the street, my 16:03 expectation is the focus will be on data 16:06 cleanliness as opposed to the outliers 16:08 that law enforcement actually should be 16:10 focusing on. Now, at this point, I would 16:13 imagine that in the comment section, um, 16:16 everybody is annoyed uh, or hopping mad 16:19 cuz that's YouTube comment section. And 16:22 I bet that I've annoyed people in such a 16:24 way that half of the comments are uh, 16:26 annoyed with me that I am softelling the 16:30 problems with Flock and that I haven't 16:31 called for the guillotining of everybody 16:35 with a badge. Uh, and then on the other 16:37 side, there's some people that feel the 16:38 exact opposite. This is a great crime 16:40 fighting tool. this will be used to deal 16:42 with the obviously real problems with 16:44 crime that we have in the United States. 16:46 Why not let people have this equipment? 16:50 And the answer to that is that I don't 16:52 believe that this actually helps the 16:55 real job of law enforcement. The real 16:57 job of the individual law enforcement 16:59 officers who are attempting to bring 17:01 justice to actual victims that have had 17:04 actual crimes committed against them by 17:06 actual criminals. Uh once you have a 17:09 tool like this, it turns into a 17:12 surveillance and supervision tool. 17:14 There's almost no way to build something 17:16 like this that is all about uh a giant 17:19 database making sure that it has the 17:21 maximum amount of data possible that it 17:24 has the maximum number of people inside 17:26 the largest possible database of faces 17:28 the largest possible understanding of 17:31 how people travel around a city and the 17:33 patterns of life that fit inside super 17:35 communities without it focusing on 17:37 collective actions rather than specific 17:40 crimes and facts. And if it's true that 17:42 our negative attention brought against 17:45 the flock cameras themselves result in 17:47 no more images being captured uh while 17:49 we still collect everything else all the 17:52 other metadata. In some ways that's 17:54 worse. Uh if I'm ever accused of a crime 17:57 because my vehicle was nearby or it was 18:00 estimated to have gone in a place that 18:01 puts me near the scene of a crime and my 18:03 other devices are somehow sort of in the 18:05 vicinity. I would actually love for 18:08 there to be exculpatory evidence from 18:10 cameras of me not being in the car or me 18:13 being under duress or being threatened 18:15 or kidnapped by a certain person. It's 18:17 very possible that the actual factual 18:19 evidence collected by cameras would be 18:21 far more helpful to me than this generic 18:24 system here. Especially in the age of AI 18:26 when this generic system here doesn't 18:28 exist just as a query system for random 18:32 records of person A or B or person 18:34 23879- 18:36 whatever this becomes a large 18:39 complicated network of nodes and 18:42 inferences in an AI model where all of 18:45 the activities of these people with 18:47 predictable patterns of life are used to 18:49 predict other people. Uh if certain 18:51 activities that you participate in line 18:53 up with certain activities that the 18:55 experts have decided are problematic, 18:57 well obviously you see where that might 19:00 potentially go. Uh you're building 19:02 police work over estimates and guesses 19:04 and statistical probabilities instead of 19:07 hard facts which like it or not cameras 19:10 could actually pick up on. Again, I'm 19:12 not defending the cameras themselves. 19:14 I'm saying that things could get bad 19:16 enough that we will wish that we had 19:18 them back. Um, and I have an example 19:20 that I think would be helpful. These 19:22 tools could easily be misused in the way 19:25 that London's uh, closed circuit 19:27 television cameras could [music] be 19:29 misused. For most of my life, um, London 19:32 has had more cameras per capita than any 19:34 other city. It's the most surveiled 19:36 state uh, when it comes to just cameras 19:39 themselves. And over the years, they've 19:41 built different systems for integrating 19:43 those cameras. I don't fully understand 19:45 where it's been at all the different 19:47 times uh of that tool or that uh that 19:50 measures uh lifetime, but I do know that 19:53 for the last couple of decades uh London 19:55 has had a problem with rape gangs. There 19:58 have been inquiries published that 20:00 really indicate that this is a systemic 20:03 problem. And interestingly, that massive 20:05 network of cameras that are still 20:07 cameras that do still locate individuals 20:11 uh that do record data that can be 20:13 looked at by human cops with their 20:15 actual eyeballs. Those constables that 20:18 were not on patrol and instead looking 20:19 at screens of information gathered by 20:22 cameras were not able to find uh or 20:25 bring to justice really any of the rape 20:28 gangs found in the inquiries. But they 20:30 were very very capable at finding all 20:33 the people that attended protests 20:35 against the rape gangs. So that is the 20:40 kind of misuse of a system that I want 20:42 us to be thinking about. And it goes far 20:44 far beyond the cameras. The law 20:47 enforcement officers that I talk to uh 20:49 definitely want more tools in their tool 20:52 bags. That's part of the reason the 20:53 T-Rex exists is to make tools for people 20:56 that need to carry life-saving 20:57 equipment. And that does include 20:59 magistrates as well as regular citizens. 21:02 [music] Now, T-Rex still has the policy 21:04 that we will not sell anything to law 21:06 enforcement officers in jurisdictions 21:08 where the private citizen cannot own the 21:10 gear. But nevertheless, a civil 21:12 magistrate is a job with a 21:14 responsibility. He carries the sword and 21:16 he doesn't carry it for nothing. The 21:19 problem is that a lot of our civil 21:21 magistrates don't need tools that help 21:23 them find aberant patterns of life. They 21:26 don't even need tools that help them 21:27 find bad guys. Uh, a lot of the times 21:29 they don't even need tools that help 21:31 them catch bad guys. If you read in the 21:34 papers, uh, you will find time after 21:36 time after time, uh, horrific crimes 21:39 being committed by people that have been 21:41 arrested 20, 30, 40 times and let go, 21:44 um, way, way too often. This is not a 21:49 system that solves that problem. 21:51 [music] 21:52 This is not a tool that addresses the 21:54 actual root issues that we have in our 21:57 communities. 21:59 Let me check my notes here. Make sure 22:01 that I covered everything that uh we 22:03 wanted to. All the potential misuses and 22:06 problems. Um yeah, no, we just barely 22:08 scratched the surface. 22:20 I will not be pushed, filed, stamped, 22:23 indexed, briefed, debriefed, or 22:25 numbered. 22:26 I am not a number. I am a free man.