OGP hosts quarterly webinars that bring together civil servants, civil society organizations, and expert partners to share knowledge and practical insights on implementing transparent and accountable algorithm policy within government.
These sessions are open to the public, creating a space where anyone interested in algorithmic governance can learn from real-world reforms, implementation challenges, and how OGP members are using the platform to advance transparency of algorithms and AI.
SESSIONS
Assessing the Impact of AI in the Public Sector (February 2026)
Individual-Level Transparency (March 2025)
February 5, 2026
About
Governments are increasingly using automated decision-making and AI-driven systems to deliver public services. As governments navigate these powerful tools, they need to understand the opportunities and risks such systems pose to the public sector now more than ever.
Assessing the impacts of AI systems is an integral part of government accountability. Broadly speaking, assessing impacts refers to methods used to evaluate the consequences of AI systems before, during, or after they are used. However, there is no universally accepted definition of what “impact assessments” cover at present. To fill this gap, this webinar explored practical tools and lessons learned about the current state of impact assessments, with a particular focus on grounding such assessments in a human rights framework.
Watch the recording
More Information
- Assess all systems, even if it means prioritizing to reduce compliance burden.
- Canada’s Algorithmic Impact Assessment tool is mandatory for every algorithm, but it calculates risk on a sliding scale. The assigned level of risk determines which notice, explanation, peer review, training, and human oversight elements are then required.
- The GobLab Public Algorithms Platform in Chile also proposes an open-source algorithm impact assessment that gives each algorithm a level of risk (low, moderate, high, and very high) and suggests personalized recommendations for different stages of the project lifecycle depending on the risks detected.
- Assess impacts throughout the lifecycle of an AI system.
- The European Center for Not-for-Profit Law (ECNL)’s Guide to Fundamental Impact Assessments encourages conducting impact assessments at all stages of the system. Moments when the impact assessment can be especially impactful should be prioritised, including when deciding whether an AI system should be developed only in specific circumstances, or even at all.
- In addition, speakers recommended conducting regular follow ups with multidisciplinary teams. As Jonathan Macdonald from the Canadian Treasury Board Secretariat explained, one-off assessments are not enough. “You can’t just have the AI and do your one-time impact and then let it go. A lot of the benefit of these tools is that they evolve over time, one should make sure that the evolution of these tools is still in line with your set of requirements.”
- Engage stakeholders meaningfully.
- ECNL’s Framework for Meaningful Engagement gives practical advice on how to engage stakeholders, including by providing them with sufficient, appropriate resources and ensuring that their input influences the product development or use in practice.
- Use transparency and bottom-up citizen feedback channels as another way to measure and track impacts.
- The Netherlands is currently developing such feedback channels through its updated action plan with three new commitments under the banner of Open Algorithms.
- Marlena Wisniak encouraged prioritizing transparency of high-impact systems when developing these feedback channels, especially in areas related to law enforcement, border control, and national security.
- Go beyond “box-ticking” compliance processes to avoid legitimizing harmful systems.
- Marlena Wisniak explained that the effectiveness of impact assessments ultimately relies on the good faith of AI developers and deployers: “You can formulate impact assessments in a way that obscures negative impacts and sets up inadequate risk mitigation measures, thus legitimizing harmful AI systems. Meaningful impact assessments are more than mere documentation artefacts. Impact assessments with strong legitimacy should include measures to address negative impacts and leave room for the possibility to not develop or deploy the assessed AI system.
- Jonathan MacDonald – Director of Responsible AI at the Treasury Board Secretariat in the Federal Government of Canada
- Olaf Schoelink – Senior Policy Officer at the Ministry of Internal Affairs and Kingdom Relations of the Government of the Netherlands
- Marlena Wisniak – Senior Legal Manager at the European Center for Not-for-Profit Law (ECNL)
- Canada’s Directive on Automated Decision Making and Algorithmic Impact Assessments
- Canada has a mandatory Algorithmic Impact Assessment (AIA) tool, which supports the implementation of Canada’s federal-level directive on automated decision-making. The Treasury Board is currently improving the tool by making it more user-friendly, adapting it to the new capabilities of generative AI, and providing more flexibility for low-impact systems in order to encourage compliance.
- The Netherlands’ approach to transparency and impact assessments
- The country’s algorithm register, which has over 1,300 algorithms published by 326 governmental organizations, records the impact assessments conducted for each system. To date, 500 different kinds of impact assessments have already been conducted, of which 304 are Data Protection Impact Assessments (DPIA), and the Netherlands has also designed its own impact assessment for human rights and algorithms.
- ECNL’s Guide to Fundamental Rights Impact Assessments
- This ECNL guide, co-created with the Danish Institute for Human Rights, was drafted in the context of the Fundamental Rights Impact Assessments. These assessments have been mandated as part of the EU AI Act, but the guide is based on the UN Guiding Principles on Business and Human Rights, which makes it applicable globally. ECNL’s guide encourages deployers to plan for typical case scenarios and worst case scenarios.
March 2025
About
In March, the Open Algorithms Network relaunched with a group of government officials discussing algorithmic transparency.
Transparency of public sector algorithms and AI systems can be achieved at two different levels. The first is a systemic one, where the overall system is made transparent, for instance via algorithm registers, technical documentation such as data sheets and model cards, or the publication of the algorithmic system’s source code, data, and/or models.
The second is an individual one, that aims to provide targeted transparency and explanations to individuals subjected to algorithmic decision-making. These practices can take various forms, ranging from providing notices that an algorithm has been used to individual explanations of a particular result. They are forms of “procedural protections” intended to respond to fairness and accountability concerns raised by algorithmic systems. They can also contribute to systemic-level transparency, by raising broader awareness about the use of algorithmic systems and AI in the public sector.