The Digital Services Act (DSA) obliges Very Large Online Platforms (VLOPs) and Very Large Online Search Engines (VLOSEs) to identify, assess and mitigate so-called systemic risks arising from the design, functioning and use of their services. However, the category of systemic risks is deliberately left to be defined through the engagement of stakeholders in an iterative process.

The DSA’s risk governance architecture gives researchers a direct and legally recognised function, including both an agenda-setting and control function. This allows them to broaden the scope of what could fall under the definition of systemic risk, supporting regulators.

However, this ambiguity also creates uncertainties. Researchers may find it difficult to understand which research gaps they could prioritise and how to effectively communicate their findings to regulators. Regulators are confronted with broad, diverse empirical and conceptual findings that may be difficult to implement. This report argues that these challenges can be addressed through a structured research agenda centred on assessing the severity and likelihood of systemic risks. By providing concrete examples of severity indicators, the research agenda demonstrates how regulators can organise and interpret diverse empirical findings, and how researchers can identify potential research gaps. We argue that no single method can adequately capture the indicators needed to assess the severity and likelihood of systemic risks. Instead, a combination of observational and experimental approaches is required.

We demonstrate how severity can be operationalised across the three dimensions (scale, scope and remediability). By mapping concrete indicators and associated methods, we show how the research agenda structures evidence on when risks linked to asymmetric amplification may become systemic.

To illustrate this approach, we use a case study of asymmetric amplification of political content (defined as the unequal distribution of political content caused by algorithmic ranking systems that surface some content more than others when compared to a neutral or alternative baseline). Although asymmetric amplification is associated with a variety of risks to civic discourse and electoral processes, there is an ongoing discussion as to whether these risks would also fall under the scope of the DSA. It is therefore a useful example to showcase what potential evidence would need to look like.

The conclusion outlines how the research agenda could be applied to other potential systemic risks and recommends a variety of potential measurements. These include long-term funding of research projects, the adoption of a harmonised and broad definition of public data, stronger collaboration to identify research gaps and more transparency by platforms and regulators regarding investigations and risk assessments.