Protecting sensitive information with AI-powered data masking

Lidingö stad—a municipality located in Stockholm County, Sweden— faces the same challenges as most other municipalities in Sweden to ensure compliance with the Freedom of the Press Act. The principle of public access to official documents is a fundamental principle of Sweden's constitution. This means, among other things, that municipal activities should be conducted as openly as possible under the scrutiny of the public and the media. The process of handling requests for public documents is a time-consuming task that must be managed promptly in accordance with Chapter 2, Section 16 of the Freedom of the Press Act. Some municipalities have therefore allocated resources solely to handle this task, or these resources (such as administrators) are found within the operations where the task is added to the existing work, which cannot be performed when requests are received. These resources, from a socio-economic perspective, could be used for more value-creating effort

Each individual request necessitates the responsible employee to perform a confidentiality assessment, reviewing all requested documents and determining what needs to be redacted. This includes manually redacting personally identifiable information (PII) and other classified material. This process is crucial for maintaining transparency and democracy .   

One of the challenges in maintaining the promptness requirement is that many of the requested documents can consist of, sometimes, up to several hundred pages, making the task of reviewing and identifying confidential information difficult. Therefore, it often requires employees to work together to ensure everything is done correctly.

Lidingö stad as part of the project “Rätt till insyn” decided to partner with IBM and ATEA to explore innovative ways to efficiently and effectively streamline its document assessment processes. Together, they embarked on a transformational journey leveraging the potential of generative artificial intelligence (gen AI).

Lidingö stad employed IBM watsonx.ai and Natural Language Processing (NLP) technologies to automatically detect and flag potentially sensitive elements across vast volumes of documentation. By implementing gen AI capabilities alongside existing tools, the initiative aimed to significantly reduce processing durations, enhance accuracy, and maintain regulatory adherence. This technological integration aligned seamlessly with Lidingö stad’s operational landscape.

Given the need for careful oversight, the Lidingö stad employees remained integral participants in the process through a collaborative approach called human-in-the-loop (HITL). HITL integrates human input into the development and use of AI systems, allowing employees to verify the AI’s automated suggestions and adjust outputs as necessary.

Over a 5-week development period, IBM conducted weekly iterations to refine and tailor functional deliveries to meet Lidingö stad’s unique needs. The collaboration produced an innovative Minimum Viable Product (MVP) powered by watsonx NLP and gen AI technologies, automating the identification and removal of PII. The result was a game-changer.

Lidingö stad was able to effectively tackle the core obstacles inherent in conventional data masking practices. What was once a manual, process was transformed into an automated, AI-assisted workflow that supports compliance with the Freedom of the Press Act.

Transitioning from purely analogue practices to semi-automatic processes has also facilitated more consistency and fostered greater institutional responsiveness. IBM will continue refining Lidingö stad’s AI models, ensuring they remain up to date, while ATEA will provide the necessary infrastructure and local expertise to integrate future developments seamlessly. In Q4, the collaborators will co-create a pilot that uses AI to digitize, identify, and collect data from disparate sources.

Both IBM and ATEA will help scale the current solution across other municipalities. Together, they will ensure the municipality continues to evolve and meet new challenges, creating a sustainable, efficient solution for public document management and data masking.

Lidingö stad views this initiative as part of a broader vision for document handling, aligned with ongoing collaboration alongside other municipalities within a Vinnova-funded project. This larger endeavor aims to establish a comprehensive framework for modernizing public document management across Sweden, with PII masking as a cornerstone. 

Lidingö stad (link resides outside of ibm.com) is a municipality located in Stockholm County, Sweden responsible for overseeing local services such as schools, emergency services, and environmental initiatives.