IAMA: Impact Assessment for Human Rights and Algorithms
You assess fundamental rights; we describe the data, how the algorithm works, bias and monitoring. For Rijkswaterstaat, we built monitoring for AI models.
To ensure the quality of its AI models in production, we provided Rijkswaterstaat with a generic package for drift and outlier detection
Rijkswaterstaat has put several AI models into production. Manual monitoring does not scale, so we developed a single monitoring package for all of them. “DSL came up with an inventive solution that we can put to good use.” Nanine Overgaauw, Rijkswaterstaat - SRE Data Science
Can this algorithm be used, and how does it affect the people it makes decisions about?
The IAMA helps public authorities assess an algorithm’s impact on fundamental rights in advance, through a discussion between the project manager, developer, legal expert and policy team. We bring the technical facts to the table and answer questions about them.
The technical questions in the human rights assessment, answered in terms a lawyer can evaluate
Purpose and data documented
What data the algorithm uses, where it comes from, what gaps and errors it contains, and which group characteristics are present or indirectly reflected.
How it works, explained
How the model reaches an outcome, which characteristics carry the most weight, and how you explain an individual outcome to the resident or employee it affects.
Bias measured by group
We test whether outcomes differ between groups and identify where those differences come from. You and your lawyer assess whether a difference is justified.
Monitoring and a shutdown plan
For Rijkswaterstaat, we developed drift monitoring for AI models in production. A shutdown plan also tells you when the assessment needs to be revisited.
The IAMA belongs in the design process, well before the algorithm is put into use
-
Purpose and rationale
Together with the project lead and policy team, we define the problem the algorithm solves and whether it can be solved without the algorithm.
-
Examining the technology
We describe the data, model, bias and explainability, and answer the legal adviser’s questions through dialogue.
-
Building in safeguards
We build in what the IAMA calls for: testing, human intervention, monitoring and a plan to stop using the algorithm.
Does our algorithm need an IAMA, and what part do you handle?
A good fit if
You’re a government body or public organisation looking to build, buy or use an algorithm or AI system that affects people, and technical questions remain open in the IAMA discussion.
Not a good fit if
You’re looking for someone to weigh the competing interests or approve the IAMA on behalf of your organisation. That decision rests with your leadership, policy team and legal team; we don’t provide legal advice.
What determines the scope
Whether it’s your own model or a purchased system, how many data sources are involved, whether a DPIA is already in place, and whether we’ll also build the monitoring after go-live.
Is an IAMA mandatory?
The government recommends the IAMA, particularly for high-risk AI and algorithms with a significant impact. Article 27 of the AI Act does, however, require a fundamental rights impact assessment for certain high-risk AI systems.
An IAMA on its own, alongside a DPIA, or as a fundamental rights assessment for high-risk AI?
An IAMA on its own
Choose this if the algorithm affects people but does not process personal data and is not high-risk AI. The Algoritmekader also recommends an IAMA for algorithms with a significant impact.
IAMA and DPIA together
Choose this if the algorithm also processes personal data. Our advice: follow the guidance from BZK and CIO Rijk, which links overlapping questions.
IAMA for Article 27
Choose this if you are a public institution planning to use high-risk AI. The February 2026 version of the IAMA has been aligned with Article 27 and identifies the relevant questions.
If you build and monitor models, you can answer the technical questions in an IAMA
Data Science Lab has worked on data and AI for medium-sized organisations and public institutions since 2016, as a Databricks Partner and Microsoft Partner. Our work includes projects for Rijkswaterstaat, ProRail and LUMC. We don’t provide legal advice.

One proposed algorithm and the outstanding questions from your IAMA are enough for an initial conversation
Stef Stoelwinder
Business DevelopmentStef is your point of contact. In our first conversation, we’ll discuss the algorithm, the data and the technical IAMA questions that remain open.
E-mail info@datasciencelab.nl
Phone +31 (0)20 3636 724
Which algorithm will you discuss in your IAMA dialogue?
Describe the algorithm, who it affects and which IAMA questions remain unanswered. You’ll receive a reply by email, or we’ll arrange an initial conversation.
Message sent
Thanks for your message. We’ll be in touch as soon as possible.
Frequently asked questions
According to the Algoritmekader, the Impact Assessment Mensenrechten en Algoritmes (IAMA) is a tool for public authorities that facilitates interdisciplinary dialogue and supports decisions about developing and deploying algorithmic systems. It applies to AI systems and other algorithms, and focuses primarily on their impact on fundamental rights.
The IAMA then serves as a reference for decision-making. It is part of your broader data governance.
Utrecht Data School at Utrecht University developed the IAMA in 2021, commissioned by the Ministry of the Interior and Kingdom Relations (BZK). It was updated in February 2026 based on user feedback and practical experience.
The new version is shorter, has a separate guidance document and has been aligned with Article 27 of the AI Act. Questions directly related to the Act are marked. According to Digitale Overheid, it remains a tool for dialogue and decision-making, not a checklist.
Before an algorithm is deployed, so that risks can be assessed in time. The Algoritmekader links the IAMA to problem analysis, design, verification and validation, and monitoring and management. It also links it to the roles of project manager, developer, legal adviser, and policy and advisory staff.
In 2023 and 2024, municipalities and the Belastingdienst were among the organisations that assessed fifteen algorithms using the IAMA. The lessons are set out in the report IAMA in actie.
Digitale Overheid says it is advisable to use the IAMA to assess the impact on fundamental rights when an algorithm has a significant impact or an AI system is high-risk. The Algoritmekader recommends it in particular for these two groups.
Article 27 of the AI Act does require a fundamental rights impact assessment for high-risk AI systems; the IAMA is a tool to help meet that requirement. Your legal adviser can assess what applies to your organisation.
According to Digitale Overheid, they have the same aim—to assess how proposed processing might affect the people concerned—but different scopes. A DPIA concerns the processing of personal data; the IAMA concerns the impact of algorithms on human rights.
Guidance from BZK and CIO Rijk maps the questions in the Model DPIA Rijksdienst to those in the IAMA, helping you avoid duplicate work. For the privacy aspects, see DPIA for AI.
Article 27 requires public bodies and private entities providing public services to assess the effects on fundamental rights before deploying certain high-risk AI systems. Cross-references to a DPIA are permitted, and the AI Office is working on a template.
The updated IAMA is aligned with this requirement. The Algoritmekader also mentions an amendment to Article 27 in the AI omnibus package. Check its status with your legal adviser and read more about the AI Act.
For the fundamental rights assessment, the Algoritmekader lists measures including testing the algorithm for bias and carrying out a justification test, regularly checking and improving quality, preparing a contingency plan for stopping the algorithm, and taking random samples in risk-based selection.
We build these measures into the model and pipeline using MLOps. To see what drift monitoring looks like, read the Rijkswaterstaat case study.
If it concerns a simple calculation rule whose owner understands how it works, or if the supplier of a purchased system already provides the technical answers. In those cases, your own team and the Rijksoverheid’s online completion tool are sufficient. The tool also uses a pre-scan to advise which assessment is needed.
If the algorithm then needs to be described to the public, read about the Algorithm Register. You can establish who decides what through AI governance.
What agreements on data, privacy and publication follow from the assessment of fundamental rights?
- Data governanceOwnership, definitions, access and data agreements that underpin every assessment of an algorithm.
- DPIA for AIThe privacy aspects of an AI application: which personal data it uses, for what purpose, and what risks and safeguards are involved.
- Algorithm registerDescribing government algorithms clearly, publishing them and keeping the information up to date.