This year has seen many changes to Big Tech businesses and products in the EU thanks to a grab bag of fresh digital legislation. Among them is the Digital Services Act (DSA), a new set of EU-wide rules that entered into full force on the 17th February 2024. The DSA's main goal, as stated by the European Commission, “is to prevent illegal and harmful activities online and the spread of disinformation. It ensures user safety, protects fundamental rights, and creates a fair and open online platform environment.” The legislation issues a variety of online safety obligations to digital services like social media, online marketplaces, search engines, and app stores.
The biggest platforms, called Very Large Online Platforms (VLOPs) or Very Large Online Search Engines (VLOSEs), have the most obligations. Those VLOPs and VLOSEs are companies like Google, Apple, Amazon, Meta, Snapchat, TikTok and others.
The DSA, specifically Article 35, states that the Commission may issue guidelines on how to mitigate 'systemic risks', among them the risks to elections. Elections are a concern for many countries around the world this year, and with the EU Parliament elections taking place on the 6-9th June, the EU is no exception. In the interest of ensuring free and fair elections,, the EU Commission has taken prompt steps to issue election guidelines. In February, the Commission published draft guidelines and opened a public consultation for feedback, with the official guidelines later published on the 26th March.
The guidelines include recommendations, amongst them:
- giving users greater control over algorithmic feeds such as recommender systems
- prevention of local-context specific risks
- prevention of hate speech or incitement to violence
- transparency mechanisms such as labelling of content, including for political advertising
- incident report mechanisms
- cooperation with external stakeholders including civil society organisations
- widening data access for outside stakeholders
- promoting official election information
- encouraging media literacy program investment
- reducing monetisation of legal but harmful content
- documentation recommendations such as post-election reviews, including publication of a non-confidential version.
These guidelines are intended to support VLOPs and VLOSEs to comply with the DSA, however they are not legally binding in themselves. Following the recommendations signals compliance whilst those who do not take instruction from the guidelines must demonstrate equivalent effectiveness. It is also relevant to note that the Commission states the guidelines “already account for forthcoming obligations” for new laws like 'the Regulation (EU) 2024/900 on the transparency and targeting of political advertising (“Regulation on Political Advertising”), as well as the forthcoming Regulation laying down harmonised rules on Artificial Intelligence (AI Act).' When those laws enter application, VLOPs and VLOSEs will have to follow those rules. Therefore, complying with these guidelines will not only signal compliance with the DSA, but also incoming laws.
What recommendations were included in the Commission Guidelines?
The Civil Liberties Union for Europe (Liberties) in partnership with the European Partnership for Democracy (EPD) gave feedback to the open consultation on the draft version of the guidelines. One of the pieces of main feedback Liberties and EPD submitted to the Commission was the need to implement “more extensive mitigation measures regarding political advertising [...] in particular, to stop the processing of all observed and inferred data in the targeting of political advertising”. This is crucial as, according to Liberties and EPD, political ad microtargeting is “particularly risky ahead of elections, as they could target swing voters and deliver tailored political messages that could even be in direct contradiction to those targeting different groups”' As such targeting involves the collection, processing and combination of personal data, including highly sensitive data, Liberties and EPD advocate cutting off that data pipeline.
While these guidelines help improve transparency about the advertiser, how much money is being spent, and even the parameters used to target the ad, the Commission chose not to include Liberties and EPD's suggestion to stop the utilisation of observed and inferred data for political advertising amongst its recommendations. While increased transparency is welcome, the problem of microtargeting still needs to be addressed.
We are pleased to see that the guidelines directly mention Liberties and EPD's paper DSA: New Risk Assessments To Protect Civic Discourse and Electoral Processes as an example of how to introduce mitigation measures for systemic risks in electoral processes, recognising many of the other recommendations made.
Liberties and EPD pointed to the risk of social polarisation, recommending that platforms “should consider not using design features that facilitate quick negative responses to opposing views”, as well as developing content and recommender algorithms that offer a “balanced information diet”. Indeed, included in the Commission guidelines are increasing user choice and control over recommender feeds to encourage media pluralism, as well as encouraging regular assessments of recommender systems and further transparency about recommender system design and functioning.
In terms of fairness, Liberties and EPD warn against giving preferential treatment to high-profile political figures, and argue that content moderation must treat all actors fairly. We see this reflected, to some extent, in the guidelines, as the Commission clearly states “that providers of VLOPs and VLOSEs show that content moderation decisions do not affect the equality of candidates or disproportionately favour or promote voices representing certain (polarised) views”. This addresses issues that arise when platforms favour metrics such as engagement over balanced civic discourse.
Other recommendations from Liberties and EPD concern inclusivity in civic spaces, specifically, content amplification and algorithm curation due to risks of “demot[ing] or fully silenc[ing] certain voices, leading to a lack of inclusivity in online discourse, and the lack of information voters receive on the viewpoints of marginalised groups on social issues discussed during the electoral campaign”. Another related risk is organised attacks against civil society, particularly when extremist content is amplified online and silences “civil society organisations and citizens willing to engage in discussions pertaining to the electoral campaign”.
The Commission guidelines note that platforms should consider:
“the impact of measures to tackle illegal content such as public incitement to violence and hatred to the extent that such illegal content may inhibit or silence voices in the democratic debate, in particular those representing vulnerable groups or minorities. For example, forms of racism, or gendered disinformation and gender-based violence online including in the context of violent extremist or terrorist ideology or FIMI targeting the LGBTIQ+ community can undermine open, democratic dialogue and debate, and further increase social division and polarization. In this respect, the Code of Conduct on countering illegal hate speech online can be used as inspiration when considering appropriate action.”
This addresses the risk of excluding different perspectives from civic discourse and asks platforms to clarify platform policies regarding hate speech. Taken together with recommendations to give users more control over algorithmic recommender feeds to encourage “media diversity and pluralism”, the guidelines address the need to foster inclusive discussions and give visibility to marginalised voices.
Liberties and EPD concluded their feedback by encouraging the Commission to consider “organising meetings on specific topics with both VLOPs/VLOSEs and civil society/experts, to exchange recommendations and best practices”. While the establishment of such a series of meetings was not mentioned in the guidelines, the Commission rightly suggests that VLOPS and VLOSEs should organise meetings “as well as establish channels of regular communication” with interested parties in electoral processes, such as civil society organisations. This type of communication should help platforms “identify risks that may require mitigation measures”. Indeed, throughout the guidelines, there is an emphasis on opening VLOP and VLOSE data access for research regarding elections, alongside continuous collaboration and engagement with institutions, independent researchers or civil society organisations to enable effective evaluations of risk mitigation measures taken.
What wasn't included?
Liberties and EPD's recommendations regarding limiting personal data collection to mitigate microtargeting political ads and its associated risks weren’t the only suggestions missing in the guidelines.
In the interest of protecting freedom of expression, Liberties and EPD warn against the over-removal of content protected by copyright claims in political messaging, which is why content moderation decisions should consider legitimate counter-claims/exceptions based on public interest purposes. This is not mentioned in the guidelines. As Liberties and EPD illustrate in their joint report, “the use of automated upload filters [...] cannot differentiate between exceptions and limitations from copyright and its infringement” leading to “platforms hinder[ing] public engagement and awareness, especially on issues where visual or audio material plays a key role in conveying the message”.
Whilst the Commission lists a variety of mitigation measures linked to Generative AI, particularly focusing on labelling and transparency, Liberties and EDP recommend that “platforms should avoid offering generative AI capabilities for the creation or enhancement of political advertisements. It poses an unnecessary risk to the electoral process”. The risk of Generative AI creating misleading or incorrect ads does not outweigh the benefits of creating a quick and cheap ad. However, the Commission only requires VLOPs and VLOSEs to “adapt their advertising systems, for example by providing advertisers with options to clearly label content created with generative AI in advertisements or promoted posts and require in their advertising policy that this label is used when the advertisement includes generative AI content”. Whilst labelling may be useful, this requires the advertiser to self-declare the ad as being AI-generated, and currently, there are technical limitations to techniques such as watermarking AI images, meaning they are likely to go undetected.
In conclusion, the European Commission's guidelines under the DSA offer a robust framework to mitigate systemic online risks associated with electoral processes. However, significant omissions, particularly concerning political microtargeting, highlight critical gaps. Despite these shortcomings, the guidelines stand as a commendable effort towards fostering transparency, fairness, and inclusivity in digital discourse, contributing to the protection of our democracies.
The real test, of course, will be in their implementation by platforms, which remains to be seen. Due to some of the DSA’s significant shortcomings, it will be challenging for civil society to access vital data and play the watchdog role that organisations like ours play. Nevertheless, we are committed to monitoring these developments closely.
Liberties is already conducting a project investigating political microtargeting during the EP elections campaign, details of which can be found here.
