0:00

Today in Security from Wired. This

0:06

episode is brought to you by ShipStation. You

0:08

know folks, I'm all about working smarter, not

0:10

harder. Running a successful e-commerce

0:12

business can be really convoluted and honestly,

0:15

you have enough going on, so I

0:17

want to introduce you guys to ShipStation.

0:20

I love using ShipStation because it makes

0:22

it so much easier to automate all

0:24

my shipping tasks and I can manage

0:26

all my orders in one simple dashboard.

0:29

Work less and ship more with

0:31

ShipStation, the innovative tool that helps

0:33

turn your shipping challenges into opportunities

0:35

for growth. Go to shipstation.com and

0:37

use code TECNUZ to sign up

0:39

for your free 60-day trial. That's

0:43

shipstation.com code TECNUZ. And

1:00

switch to an all-electric, all-climate heat

1:02

pump from Mitsubishi Electric. Learn more

1:04

about Mitsubishi Electric products at patriotair.com.

1:07

Federal income tax credit available to U.S. homeowners only for qualifying

1:09

heat pump models. Terms and conditions for tax credit are dependent

1:11

on federal and state laws and regulations and are subject to

1:13

change. Contact your tax advisor for further requirements and to confirm

1:15

eligibility for tax credits. The

1:17

mystery of AI gunshot detection accuracy

1:19

is finally unraveling. How

1:21

accurate are gunshot detection systems really?

1:24

For years, it's been a secret,

1:27

but new reports from San Jose

1:29

and NYC show these systems have

1:31

operated well below their advertised accuracy

1:33

rates. By Todd Feathers. Liz

1:37

Gonzalez's neighborhood in East San Jose

1:39

can be loud. Some

1:42

of her neighbors apparently want the whole block to

1:44

hear their cars. Others

1:46

like to light fireworks for every occasion. And

1:48

occasionally, there are gunshots. In

1:52

February 2023, San Jose began

1:55

piloting AI-powered gunshot detection technology

1:57

from the company Flox Safety.

4:00

But it is clear the system is not detecting

4:02

all gunshots the department would expect. Flock

4:04

Safety says its Raven Gunshot Detection System

4:07

is 90% accurate. Sound

4:09

Thinking, which sells the ShotSpotter system, is

4:11

the most popular gunshot detection technology on

4:14

the market. It claims a 97% accuracy rate.

4:18

But the data from San Jose and

4:20

a handful of other communities that use

4:22

the technologies suggests the systems, which use

4:25

computer algorithms, and in Sound Thinking's case,

4:27

human reviewers to determine whether the sounds

4:29

captured by their sensors are gunshots, may

4:32

be less reliable than advertised. Last

4:34

year, journalists with CU Citizens Access

4:36

obtained data from Champaign, Illinois, showing

4:39

that only 8% of the 64

4:41

alerts generated by the city's Raven system

4:44

over a six-month period could be confirmed

4:46

as gunfire. In 2021, the

4:48

Chicago Office of Inspector General reported that

4:50

over a 17-month period, only 9% of

4:52

the 41,830 alerts with dispositions that were

4:55

generated by the city's ShotSpotter system

5:00

could be connected to evidence of a gun-related

5:02

crime. Sound Thinking has

5:04

criticized the Chicago OIG report

5:06

saying it relied on incomplete

5:08

and irreconcilable data. This

5:11

week, New York City's controller published a

5:13

similar audit of the city's ShotSpotter system

5:15

showing that only 13% of

5:18

the alerts the system generated over an

5:20

eight-month period could be confirmed as gunfire.

5:23

The auditors noted that while the

5:25

NYPD has the information necessary to

5:27

publish data about ShotSpotter's accuracy, it

5:29

does not do so. They

5:31

described the department's accountability measures as

5:34

inadequate and not sufficient to demonstrate

5:36

the effectiveness of the tool. Champaign

5:40

and Chicago have since canceled their

5:42

contracts, with flock security and Sound

5:44

Thinking respectively. Raven

5:46

is over 90% accurate at detecting

5:48

gunshots with around the same accuracy

5:51

percentage at detecting fireworks. Josh Thomas,

5:53

flock safety senior vice president of

5:55

policy and communications, tells Wired in

5:57

a statement. And, critically, Raven

5:59

alerts