Risks and mitigations of releasing data

Risk analysis and

complexity in de-identifying

and releasing data.

Sara-Jayne Terp

RDF Discussion

First, Do No Harm

“If you make a dataset public, you

have a responsibility, to the best of your knowledge, skills, and advice, to

do no harm to the people connected to that dataset. You balance making data

available to people who can do

good with it and protecting the

data subjects, sources, and

managers.”

2

What is risk?What is the risk here?

3

RISK

“The probability of something happening multiplied by the resulting cost or benefit if it does” (Oxford English Dictionary)

Three parts:

•Cost/benefit

•Probability

•Subject (to what/whom)4

Subjects: Physical

5

“Witnesses told us that

a helicopter had been

circling around the

area for hours by the

time the bakery opened

in the afternoon. It

had, perhaps, 200

people lined up to get

bread. Suddenly, the

helicopter dropped a

bomb that hit a building

on the opposite side [of

the street] from the

bakery, spraying

shrapnel and debris

over the breadline”

- FirstMileGeo report on Aleppo

Subjects: Reputational

6

Subjects: Physical

7

Collectors: Physical

8

Processors: Legal

9

Risk OF What?

• Physical harm

• Legal harm (e.g. jail, IP disputes)

• Reputational harm

• Privacy breach

10

Risk to Whom?

• Data subjects (elections example)

• Data collectors (conflict example)

• Data processing team (military equipment example)

• Person releasing the data (corruption example)

• Person using the data

11

Likelihood of Risk

Low

Medium

High

12

piIHow I handle it

13

PII

“Personally identifiable information (PII) is any data that could potentially identify a specific individual. Any information that can be used to distinguish one person from another and can be used for de-anonymizing anonymous data can be considered PII.”

14

Learn to spot Red Flags

• Names, addresses, phone numbers

• Locations: lat/long, GIS traces, locality (e.g. home + work as an identifier)

• Members of small populations

• Untranslated text

• Codes (e.g. “41”)

• Slang terms

• Can be combined with other datasets to produce PII

15

Consider Partial Release

Release to only some groups

• Academics

• People in your organisation

• Data subjects

Release at lower granularity

• Town/district level, not street

• Subset or sample of data ‘rows’

• Subset of data ‘columns’

16

Include locals

Locals can spot:

•Local languages

•Local slang

•Innocent-looking phrases

Locals might also choose the risk

17

Consider Interactions Between Datasets

18

Learn From Experts

Over to you…

19

THANK YOU

For questions or

suggestions:

Responsible Data Forum

For questions or

suggestions:

Responsible Data Forum