By Chloé Berthélémy

On 17 October 2019, the European Parliament, the Council of the European Union (EU) and the European Commission started closed-door negotiations, trilogues, with a view to reaching an early agreement on the Regulation on preventing the dissemination of terrorist content online.

The European Parliament improved the text proposed by the European Commission by addressing its dangerous pitfalls and by reinforcing rights-based and rights-protective measures. The position of the Council of the European Union, however, supported the“proactive measures” the Commission suggested, meaning potential “general monitoring obligations” and in practice, automated detection tools and upload filters to identity and delete “terrorist content”.

Finding middle ground

In trilogue negotiations, the parties – the European Parliament, Commission, and Council – attempt to reach a consensus starting from what can be very divergent texts. In the Commission’s and Council’s version of the proposed Regulation, national competent authorities have the option to force the use of technical measures upon service providers. The Parliament, on the contrary, deleted all references to forced pro-activity and thus, put in line the Regulation with Article 15 of the E-Commerce Directive that prohibits obligations on platforms to generally monitor the user-generated content they host on their platforms.

Ahead of the negotiations, the European Commission was exploring the possibility to suggest “re-upload filters” instead of upload filters as a way towards building a compromise. Also known as “stay-down filters”, these filters distinguish themselves from regular ones by only searching, identifying and taking down content that has been already taken down once. This is to ensure that a content that was first deemed illegal would stay down and does not spread further online.

Upload or re-upload filters: What’s the difference?

“Re-upload filters” entail the use of automated means and the creation of hash databases that contain digital hash “fingerprints” of every piece of content that hosting providers have identified as illegal and removed. They also mean that all user-generated content published on the intermediaries’ services is monitored and compared with the material contained in those databases, and is filtered out in case of a match. As the pieces of content included in those databases are in most cases not subject to a court’s judgment, this practice could amount to an obligation of general monitoring, which is prohibited under Article 15 of the E-Commerce Directive.

Filters are not equipped to make complex judgments on the legality of content posted online. They do not understand the context in which content is published and shared, and as a result, they often make mistakes. Such algorithmic tools do not take proper account of the legal use of the content, for example for educational, artistic, journalistic or research purposes, for expressing polemic, controversial and dissident views in the context of public debates or in the framework of awareness raising activities. They risk accidentally suppressing legal speech, with exacerbated impacts on already marginalised individual internet users.

Human rights defenders as collateral damage

The way the hash databases will be formed will likely reflect discriminatory societal biases. Indeed, certain types of content and speech are getting more reported than others. The decision by the platforms to characterise them as illegal and to add them to the databases often mirrors societal norms. As a result, content related to Islamic terrorism propaganda will be more likely targeted than white supremacist content – even in cases in which the former is actually a documentation of human rights violations or is serving an awareness-raising purpose against terrorist recruitment. Hash databases of alleged illegal content are not accountable, transparent and democratically audited and controlled and will likely disadvantage certain users based on their ethnic background, gender, religion, language, or location.

In addition, re-upload filters are easy to circumvent on mainstream platforms: Facebook declared that it has over 800 distinct edits of the Christchurch shooting video in its hash database because users constantly modified the original material in order to trick automatic identification. Lastly, hash databases and related algorithms are being developed by dominant platforms, which have the resources to invest in such sophisticated tools. Obliging all other actors on the market to adopt such databases risks reinforcing their dominant position.

A more human rights compatible approach would follow the Parliament’s proposal, in which platforms are required to implement measures – exclusive of monitoring and automated tools – only after it received a substantial number of removal orders and that do not hamper their users’ freedom of expression and right to receive and impart information. The negotiating team from the European Parliament should defend the improvements achieved after arduous negotiations with the Parliament’s different political groups and committees. Serious problems, such as terrorism, require serious legislation, and not technological solutionism.

Terrorist content online Regulation: Document pool
https://edri.org/terrorist-content-regulation-document-pool/

Open letter on the Terrorism Database (05.02.2019)
https://edri.org/open-letter-on-the-terrorism-database/

Terrorist Content Regulation: Successful “damage control” by LIBE Committee (08.04.2019)
https://edri.org/terrorist-content-libe-vote/

Vice, Why Won’t Twitter Treat White Supremacy Like ISIS? Because It Would Mean Banning Some Republican Politicians Too (25.04.2019)
https://www.vice.com/en_us/article/a3xgq5/why-wont-twitter-treat-white-supremacy-like-isis-because-it-would-mean-banning-some-republican-politicians-too

(Contribution by Chloé Berthélémy, EDRi)