0:03

For the past four months, a

0:05

team of people from Elon Musk's so

0:08

-called Department of Government Efficiency has gone

0:10

from one federal agency to another,

0:12

looking at data. Very

0:14

early on, we found

0:16

that they had access to

0:18

the sensitive payment data

0:20

system or payment systems within

0:22

the U .S. Treasury. Victoria

0:26

Elliott reports on Doge and

0:29

Musk's operatives for wired. We

0:31

know from my colleagues reporting

0:33

that they have gained access

0:35

at the Social Security Administration

0:37

and we also know from

0:39

documents filed in another lawsuit

0:42

that when members of Doge

0:44

were at the Social Security

0:46

Administration that they saw access

0:48

to the SAVE database which

0:50

is run by the U

0:52

.S. Citizenship and Immigration Services

0:54

and that tracks people who

0:56

are in the country legally. Vittoria

0:59

and her colleagues also reported

1:01

that Doge operatives are cross -referencing

1:03

data from the many agencies of

1:05

the US government. Doge

1:08

also has been accessing

1:10

the USIS database, and

1:13

it appears that they

1:15

are querying it against social

1:17

security data, IRS data,

1:19

and also state voting data.

1:22

There's no one list of what

1:24

data or systems Doge has accessed.

1:26

Vitoria and other reporters are

1:28

carefully piecing together a larger puzzle,

1:31

based in part on what the government itself

1:33

is saying. This week,

1:35

the Department of Homeland Security,

1:37

DOGE, and the US Citizenship

1:39

and Immigration Services announced what

1:41

they called a comprehensive optimization

1:44

of one of the country's

1:46

largest immigration databases for enforcement

1:48

purposes. You know, while that

1:50

kind of data would normally

1:52

maybe be shared if there was

1:54

an investigation. It's not like

1:56

everything is perfectly siloed. It

1:58

does seem that that access and

2:01

that data overlay is much

2:03

greater than it has historically

2:05

been. The consistent refrain from

2:07

Musk and his associates is that

2:09

this is about efficiency. Just

2:12

listen to Airbnb co -founder and Doge

2:14

member Joe Gebia on Fox News

2:16

last month. We really believe

2:18

that the government can have an

2:20

Apple Store -like experience. beautifully

2:22

designed, great user experience

2:24

modern systems. And sure, in

2:26

theory, that sounds great. Dealing

2:28

with government tech systems can be

2:30

super annoying. There is a

2:32

real pain point there. But

2:34

there's also reasons that these data

2:37

are siloed and it is because

2:39

that's a safety measure. First off,

2:41

if all your data is in

2:43

the same place, all you need

2:45

is one really good hack from

2:47

a foreign adversary and you're into

2:49

everything. And

2:51

then there are the other risks. We

2:54

already know that the Trump

2:56

administration wants this data for

2:58

immigration enforcement. What

3:00

if they want it for more? I'm

3:02

thinking about, you know, RFK Jr.'s recent

3:04

announcement that they're going to try

3:06

and get medical records from people to

3:08

figure out autism, but I don't

3:10

know that I would want my IRS

3:12

data combined with my medical records.

3:15

for the purpose of the government to

3:17

surveil me, I think I would

3:19

find that extraordinarily scary. So

3:21

I don't think that it

3:23

is purely on the level of

3:25

saying we want every American

3:27

to feel like interacting with the

3:29

government is a pleasurable experience.

3:31

I think it is also because

3:33

these types of actions

3:35

make carrying out and

3:37

whatever presidential agenda you

3:39

have much easier, even

3:41

if it comes at the cost of people's

3:43

privacy. Today on the show,

3:46

data is power. It's

3:48

also the key building block

3:50

of a surveillance state. I'm

3:52

Lizzie O 'Leary, and you're listening to

3:54

What Next TBD, a show about

3:56

technology, power, and how the future

3:58

will be determined. Stick around. I

4:14

guess the kind of picture you

4:16

might be able to build of

4:18

someone from the data that you

4:21

and your colleagues have reported that

4:23

Doge has access to right now.

4:25

Let's say you're looking at data

4:27

sets, you're looking at systems. Roughly,

4:30

can you sketch out who

4:32

somebody is? I think

4:34

it would be entirely possible, and particularly

4:36

the more that you have to interact

4:38

with the government. the more

4:40

likely. So, for instance, say

4:42

you are a person who grew

4:44

up in low -income housing, maybe

4:47

you have student loans, maybe

4:50

you have an undocumented family

4:52

member, are a recipient of

4:54

public housing of Section 8,

4:56

Section 9, that means

4:58

that your information, your social

5:00

security number, maybe that your

5:02

household income, that's part of

5:04

a government database. If

5:06

you have a family

5:09

member who is an immigrant,

5:11

documented or otherwise, a

5:13

lot of times in the USIS

5:15

database, for instance... You're saying the

5:17

US Citizenship and Immigration Service. Yes.

5:19

So, for instance, information will be in

5:21

there on an immigrant and their extended

5:23

family, their sponsors in the country. And

5:25

then of course, your social security number

5:28

is tied to everything. It's tied

5:30

to employment. It's tied to

5:32

birth, death, medical records. There's so

5:34

much that... knowing someone's social

5:36

security number can give you access

5:38

to. I mean, that's why

5:40

hackers want it. There are

5:42

obviously populations that might be more

5:44

vulnerable than others based on their

5:46

interactions with the government or their

5:48

own legal status or immigration status.

5:51

But the reality is that everyone

5:53

is vulnerable to some degree of

5:55

having their information being combined in

5:57

a way that could give the

5:59

government a picture that maybe you

6:01

don't want them to have. The

6:04

refrain from Doge, and you heard it

6:06

earlier from Joe Gebbia, is

6:08

that siloed data and systems are

6:10

inefficient. But inefficiency

6:12

is also a form of

6:14

protection. Protection from someone

6:17

looking at what government services

6:19

you receive, whether you get

6:21

Medicare, what kind of

6:23

political donations you may have made.

6:26

There's all these other ways that

6:28

they can build a picture of

6:30

you and data on that level

6:32

can lead to discrimination across many

6:34

different things. Information about

6:36

your medical history can lead

6:39

to discrimination, information about your

6:41

sexual orientation or your, you

6:43

know, whether or not you're

6:45

divorced or any of these things

6:47

in the same way that

6:49

we wouldn't necessarily want insurance companies

6:51

to know those things because

6:53

they might use that data in

6:55

a way that's faraging more.

6:57

There's no guarantee that a government

6:59

is going to look at

7:01

that data and not use it

7:03

for something like predicting your

7:05

outcomes or deciding if you're more

7:07

likely to commit a crime

7:09

or investigating you if they think

7:11

that perhaps you disagree with

7:13

them. Traditionally, these

7:16

systems have been kept intentionally separate

7:18

from one another. And

7:20

as I understand the Privacy

7:22

Act of 1974 and the

7:24

way it's interpreted, The

7:27

employees carrying out a

7:29

lot of work around

7:31

these data sets should

7:33

not have access to

7:35

personally identifiable information, but

7:37

I wonder what your

7:39

reporting says about what

7:41

can be seen on

7:43

a granular level. You

7:46

know, reasonably, you're right. There

7:48

shouldn't necessarily be the ability to

7:50

see PII or personally identifiable

7:52

information, but... to some people who

7:54

had previously worked for DHS,

7:56

for instance, for the story my

7:58

colleague and I just did, we

8:00

were told like, yes, they're sort

8:03

of these different data sets, even

8:05

within DHS, because DHS

8:07

deals with granting you your green

8:09

card, and it also deals

8:11

with like homeland security investigations, like

8:13

possibly looking at people for

8:15

being threats to national security. So

8:17

there's the data set. around

8:19

the normal process of immigrating of

8:21

having to interview your family

8:23

members or things like that, that

8:25

is supposed to be kept

8:27

separate from the data that might

8:29

be used for enforcement. But

8:31

in reality, you know, there

8:33

are all these carve -outs and exceptions

8:35

for law enforcement. So... HSI, Homeland Security

8:37

Investigations, or ICE, can go to

8:39

other parts of DHS and say, hey,

8:41

we need access to this information

8:43

for an investigation for law enforcement purposes.

8:45

That's not always the case. Sometimes

8:48

they need a court order, but there

8:50

are carve -outs for law enforcement. According

8:52

to Vittoria, this is by

8:54

design. Normally, especially if

8:56

you're thinking about something like

8:59

IRS or Social Security Administration, things

9:01

that would contain really specific

9:03

data, it is supposed to

9:05

be an incredibly limited set of people

9:07

who can access something. It's very much

9:09

on a need -to -know level. The

9:12

general rule with data for

9:14

the government is the lowest level

9:16

necessary, so the absolute bare minimum

9:18

you need to do your job.

9:20

And that can make things really

9:22

slow. In fact, a lot

9:24

of the experience, I think, of

9:26

Americans with what they perceive

9:28

to be inefficiency in the government

9:30

is having to work through these

9:32

systems that don't all click

9:34

together because your data is supposed

9:36

to be protected by them. Listening

9:39

to you talk about data sharing, I

9:42

want to understand how formal

9:44

the agreements are among and

9:46

between different agencies. Are we

9:48

talking about somewhere there exists

9:50

a written policy saying We

9:52

agree to share this with

9:54

you, or is this somebody

9:57

in one department calls up

9:59

someone else and says, hey,

10:01

can you look up this

10:03

guy for me? So

10:05

it depends. So for instance, if

10:07

you're talking about data within an agency,

10:09

and I think DHS is a great example

10:11

because, again, it deals with sort of

10:13

the regular immigration stuff like you would have

10:16

with USIS or US Citizenship and Immigration

10:18

Services, and it also deals with like a

10:20

law enforcement component. That data

10:22

sharing across the agency would probably

10:24

be much easier. But when

10:26

you're sharing across agencies, you

10:28

have these agreements called computer

10:30

matching agreements, and you also have

10:32

these things called system of

10:35

record notices or sorens. And those

10:37

actually spell out this. Agency

10:39

is partnering with this agency and

10:41

they're going to be given

10:43

this kind of access for this

10:45

reason and so you know

10:48

if you go on the DHS

10:50

website you can see system

10:52

of records notices and you can

10:54

see every single soren that

10:56

they have with other agencies to

10:58

kind of understand what. part

11:00

of the agency is sharing what

11:03

and with whom. So with

11:05

the exception of maybe some sort

11:07

of criminal investigations, like if

11:09

we're just talking about routine access

11:11

across agencies, which definitely does

11:13

happen, these are all meant to

11:15

be documented publicly so that

11:18

if nothing else, there's a record

11:20

of what's being shared and

11:22

across which agencies. Are

11:24

they being documented publicly? I

11:27

think Doge is

11:29

a... distinct departure

11:31

from how things have normally

11:33

worked. Despite rulings, by the way,

11:35

that they should be subject

11:37

to FOIA and the Federal Records

11:39

Act. Yeah, but I think

11:41

the bigger thing too is, you

11:44

know, it's very unusual to

11:46

see an individual work across four,

11:48

five agencies at a time,

11:50

have simultaneous access to four, five...

11:52

systems across many different agencies

11:54

at a time, you know, it's

11:56

not uncommon, for instance, for

11:58

someone to start at an agency

12:00

and then get detailed out

12:03

to another agency. And we've seen

12:05

some of that with Doge.

12:07

We've seen obviously some exchange between

12:09

the General Services Administration and

12:11

the Department of Labor and some

12:13

other ones, you know, so

12:15

they are starting to do some

12:17

of that documentation. But

12:20

it's very uncommon that you

12:22

would have someone who is

12:24

accessing all of these things

12:26

simultaneously across multiple different agencies.

12:28

And we have seen that

12:30

many different times with Doge.

12:32

So there was the executive

12:34

order on March 20, I

12:36

believe, where President Trump said

12:39

there was the sort of

12:41

ending data silos executive order.

12:43

And that was really giving

12:45

Doge the go ahead to

12:47

combine some of these data

12:49

sets and systems that maybe

12:51

would not normally have otherwise

12:53

been interacting. And then

12:55

on April 5th, DHS actually

12:57

struck an official agreement with

13:00

the IRS to use tax

13:02

data to search for immigrants

13:04

for enforcement. So, you know, I

13:06

think we are seeing some

13:08

of these become public and

13:10

more formalized, but then we're also

13:12

seeing this informal mingling, which

13:14

is having people working

13:16

across these agencies simultaneously having

13:18

insight into these data sets

13:20

simultaneously where that would never

13:23

have been possible before. When

13:26

we come back, how

13:28

the consequences of sharing that

13:30

data will impact everyone,

13:32

including you. Last

13:43

week, the Washington Post reported

13:45

that the Social Security Administration

13:47

entered the names of some

13:49

6 ,000 largely Latino immigrants

13:52

into a database that it

13:54

uses to track dead people,

13:56

which effectively kind of erases

13:58

their ability to work legally

14:00

in the US, receive benefits. It

14:03

just seems to show the

14:05

kind of power that comes

14:07

with the ability to read

14:09

and write data. Yeah,

14:11

I think that's true. And

14:13

I think it's very telling,

14:15

for instance, that some of

14:17

the earliest people hired into

14:19

Doge were not people with

14:21

like extensive government experience. They

14:24

were technical people, often young,

14:26

you know, Silicon Valley love sort

14:28

of young scrappy people who

14:30

are going to work long hours.

14:32

They really emphasize people who

14:34

had a specific technical skill set

14:36

because they wanted people who

14:38

could go in. and access this

14:40

data and play with it,

14:42

analyze it. We're still finding out

14:44

what they're doing with it. But

14:47

if our reporting is any

14:49

indication, it seems like a big

14:51

goal is to be able

14:53

to combine it across agencies in

14:55

a way that we've never

14:57

seen before. There

14:59

was so much reporting, including by

15:01

you and your colleagues, in the

15:04

early days of Doge where It

15:06

was just becoming clear that they

15:08

had access to this and that

15:10

and this agency, that agency. Is

15:13

it fair to say

15:15

that now we are

15:17

turning from access to

15:19

intention, access to the

15:21

application of this data?

15:25

Yeah, I do think that

15:27

that is what we're seeing,

15:29

right? Because initially, when we

15:31

were seeing them access all these

15:33

different systems, You know,

15:35

we had sort of inklings. Again,

15:37

you know, the lawsuits around the

15:39

Treasury indicate how that might have

15:42

been used around USAID. We

15:44

have a sense that they

15:46

were looking for employee details at

15:48

OPM or the Office of

15:50

Personal Management, you know, to be

15:52

able to conduct these mass

15:54

firings. We sort of had early

15:56

inklings based on the actions

15:58

that were taking place at those

16:00

moments. But I think there

16:02

wasn't necessarily a

16:05

clear sense of what

16:07

was going on

16:09

with this sort of

16:11

wide ranging access. And

16:14

now it's only starting to become

16:16

a bit more clear. And I

16:18

think, again, that the immigration space

16:20

is where we're seeing it first.

16:22

That doesn't mean it will be

16:24

the only space. I think it

16:26

is, again, just one of the

16:28

more pressing parts of the president's

16:31

agenda. There's another

16:33

story that you and your

16:35

colleagues have worked on

16:37

which talks about building an

16:39

immigration OS, basically a

16:41

surveillance platform for a lack

16:43

of a better term,

16:45

where ICE would work alongside

16:47

the company Palantir, which

16:49

has been an ICE contractor

16:51

for some time. Can

16:53

you explain what that might

16:55

do because it feels incredibly

16:57

all -encompassing? So it was

17:00

interesting actually because we had

17:02

people tell us that there

17:04

were these efforts to combine

17:06

data across DHS in a

17:08

way that was sort of

17:10

had never really happened before

17:12

and then, you know, shortly

17:14

after my colleague Caroline published

17:16

that story. So it's unclear

17:18

if that sort of immigration

17:20

enforcement OS is what people

17:22

referring to or if there

17:24

are other initiatives that we

17:26

aren't yet aware of that

17:28

are also part of this. But

17:32

again, it's combining information

17:34

from USIS. And

17:36

again, ICE particularly, and

17:39

particularly within DHS, there has sort

17:41

of been a culture of more

17:43

information sharing just because these are

17:45

both nested within DHS. So that's

17:47

a little easier to do in

17:49

terms of scooping up all that

17:52

data than it would otherwise be

17:54

to go across an agency. But

17:56

I think it is concerning

17:59

because, again, one of the things

18:01

that experts have pointed out

18:03

to us is, my name

18:05

is Victoria. It's not a very

18:07

common first name in this country. It's

18:09

Italian. And the number

18:11

of times that I have accidentally been

18:13

entered into a system is Victoria. is

18:16

innumerable. When I

18:18

went to the DMV the first time to

18:20

get my driver's license when I was 16,

18:22

I had the nice lady at the DMV

18:24

tell me I'd spelt my name wrong and

18:26

helpfully correct it for me and then I

18:28

had to start the process again. Yeah, my

18:30

name has an apostrophe. Computers do not know

18:33

what to do with me. Exactly. So

18:35

I think the thing that we

18:37

should really be concerned about is this

18:39

sort of idea that like data

18:41

is valuable. matching different data

18:43

sets across different agencies and

18:45

obviously like something like a social

18:47

security number hopefully helps with

18:49

that. But these things are all

18:52

imperfect. You know, they

18:54

are liable to have clerical errors

18:56

that could really result in harmful

18:58

things. I mean, if you think

19:00

about the situation with Kilmar of

19:02

Brego Garcia, like the government

19:04

lawyer who was then dismissed admitted

19:06

essentially that it was a mistake

19:09

and I think we should Really

19:11

not as much as these data

19:13

sets and the systems are incredibly

19:15

powerful and they need to be

19:17

treated with care and protected. I

19:19

also don't think we should overestimate

19:21

their accuracy. There are

19:23

errors and problems in government

19:25

information all the time and.

19:27

The idea that somehow we

19:30

would be firing lots of

19:32

employees, removing humans from the

19:34

loop of some of the

19:36

most sensitive systems and processes,

19:39

and then combining data across

19:41

agencies without necessarily being

19:43

totally sure about who that

19:45

could target, I think

19:47

leaves so much room for

19:49

error. Immigration attorneys were

19:51

sent letters saying that they

19:53

had to self -deport. I'm sure

19:55

that was probably a clerical error

19:57

that their name was in

20:00

the system in some capacity, probably

20:02

associated with one of their

20:04

clients. And whatever tool

20:06

that DHS was using to

20:08

send out that message just didn't

20:10

pick up on that. And

20:12

if you can think of that

20:15

at a massive scale, even

20:17

small amounts of errors

20:19

could cause irreparable harm to

20:21

many people. We

20:25

have been talking largely about

20:27

immigration. Is it

20:29

too bold a question to say, this

20:32

might be starting with immigration, but

20:34

is that where it ends? I

20:37

think history and

20:39

other countries are

20:41

great examples for

20:44

the answer to

20:46

that question. I

20:48

think about what we're seeing in

20:50

Turkey right now. favored

20:52

presidential contender to

20:55

run against Erdogan is

20:57

currently facing a

20:59

tax investigation. I

21:01

think immigration is something that's sort

21:03

of at the fore of this administration's

21:05

agenda. It's also been

21:07

something that has historically been

21:09

popular with MAGA supporters. But

21:12

the reality is that once you have

21:14

access to the sensitive data in the IRS,

21:17

you can investigate an American citizen

21:19

as easily as you can investigate

21:21

an immigrant. And once you're

21:23

breaching the norms around how

21:26

that data should be used

21:28

to begin with, I think

21:30

there is very little friction

21:32

in pushing it further. And

21:34

that doesn't mean that we've

21:37

seen them do that yet.

21:39

But that certainly could be

21:41

conceivable with the access that

21:43

they have. All

21:45

of this brings to mind

21:47

something that billionaire Larry Ellison, the

21:49

oracle founder and prominent Trump

21:51

backer, said last year. That

21:54

AI -powered surveillance systems would

21:56

usher in an age where,

21:58

quote, citizens will be

22:00

on their best behavior. I

22:03

think the thing is that, like, we

22:05

have seen in

22:08

Silicon Valley a willingness

22:10

to break rules. a

22:12

willingness to allow their tools

22:14

to be used in anti

22:16

-democratic ways, a

22:18

willingness to sort of push the limits

22:21

and apologize later. Move fast and

22:23

break things. Yeah. And

22:25

I think when you are

22:27

doing that on a government

22:29

level, first off, 90 %

22:31

of startups fail. I don't know

22:33

that we can ride out the

22:35

eventuality of the US government failing. I

22:38

don't love that. And I think the

22:40

real thing is this. When

22:42

people talk about AI or any

22:44

of these tools, what they're really talking

22:46

about in many ways is efficiency. That's

22:49

why they're so excited to

22:51

introduce tools like GSAI into the

22:53

General Services Administration, which my

22:55

colleagues had people there tell them

22:57

was not particularly better than

22:59

an intern. But it's

23:01

this sort of idea that

23:04

Yes, there will be surveillance, but

23:06

also this sort of idea

23:08

that the thing can run on

23:10

its own. And

23:12

so I think

23:14

even before we pick

23:16

apart what Doge

23:18

is doing, the

23:20

really big question that I don't

23:23

think they've managed to really answer

23:25

is efficient for whom and for

23:27

what purpose. And yeah, does

23:29

combining all these data sets make

23:31

immigration enforcement quote unquote more efficient?

23:33

Yes, it makes it more automated.

23:35

It makes quote unquote targets easier

23:37

to find. Does that make it

23:39

accurate? Does that make it

23:42

just? When we

23:44

talk about so much

23:46

of this, it's that

23:48

justice, fairness, dealing

23:51

with human beings and

23:53

creating systems that cater

23:55

not just to the

23:58

way Silicon works, but

24:00

systems that truly work

24:02

and cater to everyone

24:04

are not going to

24:06

feel shiny and

24:08

Apple Store efficient. They're probably

24:10

going to be slower and

24:12

clunkier, but that's okay because

24:14

not everything needs to run

24:16

like a business. Not everything

24:18

is a market cap. Victoria

24:24

Elliott, thank you so much for your

24:26

reporting and for talking with me. Yeah,

24:28

thank you for your time. Victoria

24:32

Elliott a reporter for Wired and

24:34

that is it for our show

24:36

today. What Next TBD is

24:38

produced by Patrick Fort. Our show is

24:40

edited by Evan Campbell. Slate is run

24:43

by Hilary Fry and TBD is

24:45

part of the larger What Next family. And

24:47

if you're looking for another great

24:49

Slate show to listen to, check

24:51

out Thursday's episode of What Next.

24:54

A union leader makes the impassioned

24:56

case that we all go on

24:58

strike. All right, we

25:00

will be back on Sunday with another

25:02

episode about what the heck is going

25:04

on with the government weather services. I'm

25:07

Lizzie O 'Leary. Thanks so much for listening.