Notes from the КРС roundtable on the Digital Services Act and Bulgaria's presidential elections

The Communications Regulation Commission convened institutions, platforms and civil society this week to ask a practical question. How should Bulgaria use the instruments of the Digital Services Act to limit systemic risks to the presidential vote, and how should the many bodies with a role in that effort actually work together. BFMI was glad to take part, alongside the Central Election Commission, the Council for Electronic Media, the data protection and cybercrime authorities, several ministries, our colleagues at Sensika, the Association of European Journalists and the Center for the Study of Democracy, and representatives of Meta, TikTok, Google and X.

Much of the discussion concerned itself with disinformation tactics and logistics, such as who flags what, to whom, on what timeline, and what are the systemic risks that Bulgaria, but broadly any upcoming election, faces. There was clear common ground that the threats are evolving faster than the arrangements built to meet them, with AI-generated content, automated networks and increasingly sophisticated manipulation techniques all raising the difficulty. Additionally, there were the very pertinent questions on funding of political campaigns online and tracing the provenance of this funding, specifically in relation to FIMI. The platforms, for their part, described continual revision of their policies and detection systems, which is welcome and necessary.

BFMI’s raison was more fundamental and we were glad to hear it seconded from several points around the table. Cooperation presupposes a shared object. Before institutions, platforms and researchers can exchange information usefully, they have to be describing the same thing when they do. On coordinated inauthentic behaviour, which is the heart of what electoral monitoring is meant to catch, no such shared description exists.

The term is not defined anywhere in binding EU law. It began as a platform coinage, introduced by Facebook in 2018, and each company now applies its own version. Meta anchors its definition on fake accounts. TikTok reaches considerably further, explicitly covering networks of authentic accounts acting on behalf of an undisclosed principal. X and Google maintain no distinct category at all, folding the conduct into general rules on spam and manipulation. The consequence is that the same operation, real influencers paid to push a campaign narrative, for instance, can constitute a violation on one platform and remain entirely permissible on another. No regulator, and no flagger, can hold the field to a common measure, because there is no common measure to apply.

This is compounded by another conflation. Manipulation and disinformation are not the same problem, though they are mutually reinforcing. Disinformation concerns the falsity of a claim, whether something circulating is true. Manipulation concerns the machinery of its spread, the coordinated behaviour that lends a claim the appearance of being organic and widely held when it is neither. A narrative can be entirely accurate and still be manipulated into false prominence, and a coordinated network can manufacture the impression of consensus without telling a single provable lie. The two problems require different instruments and treating them as one is how the behavioural side slips out of view.

In an electoral campaign this matters substantially more, because the harm to a vote comes less from any particular false statement than from a manufactured sense of an engineered impression of the national mood that shapes how people understand their own society and, in turn, how they vote. That is a problem of behaviour and of coordination, and the reflexes of fact-checking, trained on whether content is true, are not adapted to catch it.

The implication for the roundtable's own goal is direct. Elections need a dedicated understanding of coordinated inauthentic behaviour, scoped to campaign periods and built around an agreed set of tactics that appear consistently and systematically in this new arsenal of electoral campaigning. General definitions can reasonably differ across platforms, since their architectures differ, but electoral manipulation needs a common floor, because the democratic process being protected is singular whatever platform a voter happens to be on. The Digital Services Act already obliges the large platforms to mitigate systemic risks to electoral processes, yet it does not define, in its binding provisions, what electoral manipulation concretely consists of. A dedicated standard would give that obligation a shared object.

BFMI left the discussion encouraged by the seriousness of the room and by the appetite for cooperation. The next step, in our view, is the one that makes all the others possible. Agree what we are jointly trying to monitor, prevent and act upon, and the mechanisms for doing so become buildable. Leave it unresolved, and even the best-intentioned coordination risks exchanging information that does not line up.

Our readers read next:

Featured