Anonymising open documents

Municipality of Utrecht

We would like to make information publicly available to the city in an efficient and responsible manner. We do this by automatically recognising and anonymising privacy-sensitive information in documents.

Last change on 12th of July 2024, at 9:28 (CET) | Publication Standard 1.0

• Publication category: Impactful algorithms
• Impact assessment: DPIA, ...
• Status: In use

General information

Theme

Organisation and business operations

Begin date

-

Contact information

[e-mail]

Responsible use

Goal and impact

Automated recognition and anonymisation of privacy-sensitive information in documents

Considerations

Results Uthiek Assessment: automated recognition and anonymisation of privacy-sensitive information makes our document disclosure process better and faster. This contributes to the Privacy value because it reduces the risk of a data breach. This is also because an employee has the last word with every document. By carrying out disclosure better and faster, we can better provide residents with information and be transparent. This increases the information position of residents towards the municipality, and contributes to the Power Relations value.

Human intervention

The algorithm (with a direct impact on Utrecht residents and their living environment) only serves as a source of information for an employee, who takes the appropriate next step based on this information and the context of the situation. This principle is included in our Utrecht framework Internal Guidelines for Algorithm Application.

Risk management

The Utrecht data scientists, Information and Process Advisors (IPA) and Decentralised Information Security Officers (DISO) within the Utrecht municipality have a controlling and monitoring role in the application of the algorithm, in addition to the employees directly involved. By doing so, we want to prevent the following risks from occurring:

- data leak by making personal data of Utrecht residents publicly available

- algorithm takes a decision completely independently, without the intervention of an employee and without interpreting the relevant context of the situation

- 'function creep' regarding data in the application of the algorithm, causing the algorithm to give a distorted picture of what is going on. This can occur because the same type of data is used when applying the algorithm, and the algorithm uses that as confirmation of what comes out as information

Impact assessment

• Utrechts Uthiek Assessment
• Data Protection Impact Assessment (DPIA)

Operations

Data

- existing open documents with 'privacy proof' employee choices

- fictitious documents with fictitious personal data