Practical experiences of AI-supported masking

Lantero interviewed Therese Forsberg, an investigator at the Department of Administration in Uddevalla Municipality. Therese works with redaction of documents in response to requests for public records, and she has been using Lantero Redact over the past months — receiving AI-based support for assessment and redaction. Below is a slightly shortened version of the interview. (Video version is in Swedish)

Interviewer: Uddevalla is a municipality with around 60,000 residents. When it comes to requests for public records, what kind of volumes are you dealing with?

Interviewee: It varies. It depends on what’s happening in the organisation. When incidents occur that lead to deviations or Lex Sarah cases, the volume increases. We also have some media outlets that submit weekly requests for all incoming records from the past week. That’s the case for municipalities across Sweden — some outlets do this continuously. So the amount can fluctuate a lot, especially if serious cases have come in.

Interviewer: To what extent is this possible to plan for?

Interviewee: Some parts are always manageable, but it becomes difficult when large volumes come in — sometimes thousands of documents. We don’t have a dedicated person working on this full-time, so our department has to share the workload.

How the process used to work

Interviewer: What did the routines look like before?

Interviewee: We did everything the old-fashioned way. We printed out the documents and redacted them manually using Tipp-Ex. Then we copied and scanned them before sending them off. Adobe has some tools, but they haven’t been reliable. You could sometimes lift off the redaction digitally, so we always had to print and scan everything anyway. It was time-consuming and difficult to manage when working remotely.

How the work is done now

Interviewer: What does the routine look like now?

Interviewee: It’s much faster. With the redaction service, we can mark what we want to redact digitally and save it directly. We avoid all the printing and scanning, which saves a lot of time. I also feel that we have better oversight of the documents and the process.

Interviewer: One idea with the AI support is that more people could participate in the work by accepting or rejecting suggested redactions. Have you started expanding that responsibility?

Interviewee: Not yet. We’ve involved some colleagues, but they have the same level of knowledge as we do. So for now, the responsibility remains within our department.

Model training and new updates

Interviewer: You recently received an updated version of the service. Have you had a chance to test the new capabilities?

Interviewee: Very briefly, but what I saw looked good. I need to test it more before I can say anything definite.

Interviewer: Do you think the assessments look similar across different municipalities?

Interviewee: Yes, I think so. We all work with the same types of documents and the same regulations. The goal is always to protect the individual and avoid revealing personal data. That should lead to similar approaches to what needs to be redacted.

User experience of the service

Interviewer: Any final reflections?

Interviewee: The service has been easy to use. We’ve found it user-friendly and free from issues. It has worked throughout the entire test period, which has been very valuable since we’ve had unusually large volumes of cases recently.