How to Protect Data

How to protect data

Terence SpiesCTO

Voltage Security

© 2014 Voltage Security, Inc.

The Big Data Security Paradox

More analysis value

More data

More attacker value


How to resolve?


Big Data Reference Monitors


Assumptions


In reality….


Cryptography

• Assume the reference monitor fails

• Make some careful mathematical assumptions

• So why isn’t every security problem solved?


-rJ BobI \ c f y: ]

A l ice Q u a n mKey Key

111,o Eve -

C.:E(M,.JQ .' Alice Bob

e,.

-' ·Ciphertext = Encrypt (Ptaintext, Key)Plaintext =Oec1yp1(Ciphe11ext. Key)Ir/!! o•=.,- .

r,..."*-0

-·· AliceCltfp!IO'l"-tloft

. l -

mess,&.ge to Bob 1?

,

Alice

Bob

1...A

cipher text

1{a •channel

9 1 ' n truder

receiver1der

·rpri\oillit kfY

OIAlic&

fpublic key

o1Alice

, 1 ......mes.sage

445 x 444 - unlvers&-revl. .

r·-····-·-·-·-·-·-····,,.._. .._..,_..,_,,_,,_,,_ l

! =- =:.. !+ ,...... l o l!ok' !

' * - " ',.,...luillal(rjft"' . ...._.- *-,,,.,...--,.....

, !

(.;\,

lx0Ak3 . . . ,o$2R

hel10Alice

tAl.JC'£

L \ t.

ftOR

X' ,.

'"ll•tmt t

r

--

Alice Bob

[r


Data Protection on Model


What’s different?

• Keys work differently

– They might not be shared at all

– Key changes might be more expensive

– Transitory data, key rotation = rollover

• Data works differently

– In communication, channel is a wire

– In storage, might want to release properties

• We need to consider the stack…


The Stack


The Encryption Stack


Current Hadoop Efforts


• JIRA-6134, JIRA-10150• Per-directory file level encryption

- Enable per-user access control files- Easy to use, good start at protection

Distributed Big Data

• The stack becomes a matrix

– Layers per machine

– How do we protect across?

• The answer is de-identification of data

– Replace the data with a “good enough” substitute

– Minimal value to attackers

– Retain analytic value


1:1 De-identification

Name SS# Credit Card # Street Address Zip

Kwfdv Cqvzgk 161-‐82-‐1292 5421 98BT QIRP 6981 2890 Ykzbpoi Clpppn 34984

Veks Iounrfo 200-‐79-‐7127 5587 08MG KYUP 0139 406 Cmxto Osfalu 23495

Pdnme Wntob 095-‐52-‐8683 5348 92VK DEPD 2829 1498 Zejojtbbx Pqkag 93483

Eskfw Gzhqlv 178-‐17-‐8353 4929 43KF PPED 4379 8261 Saicbmeayqw Yotv 02489

Jsm Tbluhm 525-‐25-‐2125 4556 25ZX LKRT 1830 8412 Wbbhalhs Ueyzg 94388

Name SS# Credit Card # Street Address Zip

James Po9er 385-‐12-‐1199 5421 9852 8235 6981 1279 Farland Avenue 77901

Ryan Johnson 857-‐64-‐4190 5587 0806 2212 0139 111 Grant Street 75090

Carrie Young 761-‐58-‐6733 5348 9261 0695 2829 4513 Cambridge Court 72801

Brent Warner 604-‐41-‐6687 4929 4358 7398 4379 1984 Middleville Road 91706

Anna Berman 416-‐03-‐4226 4556 2525 1285 1830 2893 Hamilton Drive 21842


De-identification Taxonomy


Cautions

• Even de-identified data can be re-identifiedin some cases

• Allows correlation, but typically not datarecovery

• Some regs (HIPAA, etc.) containre-identification on guidelines


De-identification Methods

• Create a 1:1 map in a database

– Simple in theory

– Problematic in many implementations

• Replication

• Encryption of database (search)

• Hash function

– Almost guaranteed to fail

– “One-way function” != you think it means


One idea from 1997…

“Ciphertext (data in encrypted form) bears roughly the same resemblance to plaintext(data in its original form) as a hamburger does to a T-bone steak.”


The FPE problem

128 bits

AES

128 bits

CCN (49 bits)


One solution….


In 2002, a new tool…


Basic idea: Luby / Rackoff

• Split the plaintext into a Left and Right half

• Run a series of rounds like so:

Leu ’ = Right

Right’ = Leu + F(Right)

• This Feistel network is almost magic– Probably secure (with the right F)

– Reversible


FPE network


Standards

• Payment space

– PCI-DSS Tokenization guidelines

– X9.119 part 1 and 2

• Health care

– HIPAA and relatives

• FPE methods

– NIST SP800-38G

– X9.124


How to Protect Data

Software

Transcript of How to Protect Data