Study: Social media platforms fail to remove illegal hate speech

Preliminary results from an EU-supported project shows that most reported hate speech on social media is either not removed or left unanswered. Researchers from the IT University of Copenhagen are a part of the study and responsible for collecting and analysing the data on hate speech in Denmark.

Computer Science Department Digital Design Department Research algorithms ethics social media

Written 3 July, 2024 09:57 by Mai Valentine Kristensen

Since January 2023, the EU-supported project 'Safenet: Monitoring and Reporting for Safer Online Environments' has monitored hate speech across 19 EU countries, including Denmark. The project's goal is to gather data, analyse, and provide insights into how social media platforms moderate hate speech to address the issue.

"We see that platforms have some procedures and processes for removing harmful content, but it is clear that they are not always effective. There are many people using these services, including children, therefore it is important that there is a certain amount of protection," says Associate Professor Christian Hardmeier from ITU.

Hate speech, and content that incites violence on social media, is illegal in the EU. Platforms like Facebook, TikTok, and X (Twitter) are required by law to remove it. But preliminary results from the study indicate that – except for TikTok – most reports of hate speech are left unanswered and an even smaller percentage is removed.

X Introduces EU Reporting Option

In Denmark, Associate Professors Luca Rossi and Christian Hardmeier along with Research Assistant Peter Brunsgaard Trolle from ITU are responsible for collecting data on how Facebook, X, and TikTok moderate hateful content.

In their latest report spanning from January 2023 to June 2024 they have reported 1130 instances of hate speech across the platforms Only 247 posts or comments were removed. TikTok is fastest to respond, with about half of the reports leading to content removal, whereas X has the longest response time and Facebook the lowest removal rate.

While X and Facebook perform similarly in the latest report this was not the case at the start of the project. Since January 2024, X has improved in responding to reports and removing harmful content. This improvement is due to a new reporting feature that allows users to report content under EU legislation instead of general guidelines.

"There is a big difference depending on which reporting function you use. If you use the general one the content will be completely removed from X – that is, if they agree with the report. If you use the EU function, it will only be removed in the relevant country. You can still find the content if you use a VPN or are in another country," explains research assistant Peter Trolle.

Before January 2024, X’s removal- and response rate was extremely low compared to the other platforms, and they still remain so if the researchers report content under general guidelines.

"Having two reporting options place a significant responsibility on users to decide whether the content is illegal under EU law or not, as the figures from X show that the two functions trigger different processes within the platform, " says Christian Hardmeier.

Few Danish-Speaking Moderators

In February 2024, the EU regulation 'Digital Service Act' (DSA) came into effect, aiming to regulate online platforms such as social media. As part of the regulation, online platforms and search engines with more than 45 million active users in the EU per month are required to publish two annual transparency reports on content moderation.

The latest figures from these reports show that X and LinkedIn have zero Danish-speaking content moderators, Facebook and Instagram have six combined, while TikTok has 27. This may partly explain why TikTok performs better than any of the other services.

Despite the different numbers of moderators, the researchers at ITU see a general trend across platforms. When they report extreme content that incites violence it is more likely to be removed, whereas more subtle content often remains unresolved or is not removed at all.

"Even though the law states that certain forms of hate speech are illegal, applying these criteria in practice is difficult because different people reach different conclusions. Furthermore, people are often subtle and ambiguous in their comments, making it even harder," says Christian Hardmeier.

Peter Trolle adds, "It requires moderators who are familiar with the language and culture of the relevant country, and the platforms are run by multinational companies."

Across the 19 EU countries participating in the Safenet project, 14,006 cases of hate speech have been reported from platforms such as Facebook, Instagram, X, YouTube, and TikTok from January 2023 to June 2024. 4,158 reports have led to content removal, corresponding to 30 percent. In the overall report, TikTok remains the platform with the fastest response time and highest removal rate while Instagram and YouTube, neither of which is included in the Danish data collection, are at the bottom.

The Safenet project will continue until the end of 2024 and aims to report 20,000 cases of hate speech in various languages. The reports are shared with the EU and the platforms involved in the study to highlight issues in regard to monitoring this type of content on social media platforms.

Find the full reports here.

Further information

Theis Duelund Jensen, Press Officer, +45 25 55 04 47, thej@itu.dk