The Cost of Weaker Moderation

In June, European Parliament lawmakers passed the Return Regulation, a legislation which aims to speed up the removal of third-country nationals from the European Union.

As the vote concluded, far-right Members of the European Parliament (MEPs) began chanting “send them back” in the chamber. Associated with anti-immigration rhetoric, the language was widely condemned as dehumanising and prompted criticism from across the political spectrum. MEP Abir Al-Sahlani stated she had “never felt as unsafe in [the European Parliament] as after the voting about the Return Regulation” and that the event had crossed a line.

Following the incident, she became the target of online abuse after far-right MEPs, notably Finnish MEP Sebastian Tynkkynen, posted about her intervention, with users calling for her to “go home”.

CCDH’s research suggests this abuse forms part of a wider trend. In its Safety Off report, it found that hateful comments towards US Members of Congress nearly quadrupled in the six months after January 2025 when Meta relaxed its content moderation policies –including by lifting restrictions on topics such as immigration and gender.

The below report supports previous research by analysing Facebook and Instagram comments under posts by MEP Abir Al-Sahlani and MEP Sebastian Tynkkynen about the EU Return Regulation. CCDH researchers found that over 40% of the comments analysed were coded as hateful or anti-migrant.

The abuse MEP Al-Sahlani faced online reflects a broader trend both in politics and online discourse. Social media platforms’ policies continue to allow hateful and discriminatory content to remain online and, in some case, be amplified. At the same time, extremist politicians are exploiting these platforms to disseminate hateful rhetoric and mobilise harassment against political opponents.

CCDH calls on the European Commission to investigate whether Meta’s policy changes comply with the Digital Services Act. Given CCDH’s previous findings about increasing online hate towards U.S. Members of Congress after Meta’s policy changes and the scale of the hate towards MEP Al-Sahlani exposed below, more transparency is needed from Meta about why their policies were changed, how content moderation is being conducted on the platform, and how this is affecting European users thus far. Online abuse of politicians, particularly the gendered, sexualised and threatening abuse disproportionately directed at women, can intimidate elected representatives, silence voices, deter participation in public life and ultimately weaken democratic debate. When platforms make decisions that affect the health of Europe’s information space and democracy, there needs to be accountability.

Online Abuse around the EU Return Regulation

CCDH analysed 16,164 Facebook and Instagram comments under posts by MEP Abir Al-Sahlani and MEP Sebastian Tynkkynen about the EU Return Regulation.[i] The analysis found that across the posts, 6,986 comments (43%) were coded as hateful or anti-migrant.

Al-Sahlani posted about feeling unsafe after right-wing lawmakers chanted “send them back” in the European Parliament and described the rhetoric as racist.[ii] Tynkkynen’s related post mocked Al-Sahlani and other MEP’s response and encouraged support for the “send them back” message.[iii]

Category	Number of comments	% of comments
Anti-migrant	5,454	34%
Hate	1,532	9%
Not relevant	9,178	57%
Total	16,164	100%

We grouped comments into three mutually exclusive categories:

• Hate: Direct attacks, dehumanising language, harmful stereotypes, or threats targeting people based on protected characteristics such as race, ethnicity, religion, or sex.
• Anti-migrant: Abuse towards individuals or groups because they are migrants, including comments telling people to “go home,” “go back to your country,” or “send them back.”
• Not relevant: Comments that did not fit either category.

CCDH used GPT-4.1 to classify comments into the three categories. To assess performance, we compared the model’s labels against a human-labelled validation sample of 500 comments. The model achieved an F1 score of 0.85, indicating strong agreement with human coding while allowing the analysis to be scaled across the full dataset of 16,164 comments.[iv]

9% of comments analysed contained hate likely to violate Meta’s policies

In total, 1,532 comments, or 9% of all comments analysed, were coded as hate.

Our definition of hate was based on Meta’s Hateful Conduct policy, which prohibits attacks on people based on protected characteristics, including race, ethnicity, religion, sex, gender identity and national origin.[v] 

Under these rules, content may violate Meta’s policies when it includes threats, dehumanising comparisons, calls for harm, slurs, or harmful stereotypes targeting protected groups. Yet many of these comments appeared under the politician’s posts.

Examples of hate included:

• “Kill the Muslims cockroaches” (Comment under Tynkkynen’s Instagram post)
• “I would prefer reopening of Auschwitz, but yeah send them back should work too” (Comment under Al-Sahlani’s Facebook post)
• “That bitch LOOOVES the Chimp cock” (Comment under Tynkkynen’s Instagram post)

251 hate comments contained gendered abuse

Within the hate-labelled comments, 251 comments, or 16% of all hate comments, were coded as gendered abuse. Most appeared under Tynkkynen’s Instagram post, which accounted for 204 of the 251 gendered-abuse comments.

Gendered abuse is abuse that targets women and girls through misogynistic, sexualised, or gender-based attacks.

Examples included:

• “Fuck you cunt, send them back” (Comment under Al-Sahlani’s Instagram reel)
• “STFU WOMAN..then go get slashed or raped…” (Comment under Tynkkynen’s Instagram reel)
• “This is why women have no right in politics” (Comment under Tynkkynen’s Instagram reel)

34% of comments analysed were anti-migrant

A further 5,454 comments, or 34% of all comments analysed, were coded as anti-migrant. These comments abused individuals or groups because they were migrants, portrayed migrants as threats, invaders, or criminals, or told people to “go home”, “go back to your country”, or “send them back”.

Of these, 2,673 comments featured some version of the phrase “send them back”, showing how rhetoric used in parliamentary debate can spread across social media.[vi]

Examples of anti-migrant comments included:

• “She has to be send back with them all” (Comment under Tynkkynen’s Instagram post)
• “They broke into our country’s and expected everything free changing our culture attacking the natives not trying to intergrade. OVER we want our country’s back” (Comment under Al-Sahlani’s Facebook post)
• “Can we vote to send those idiots back with them ? SEND THEM BACK.” (Comment under Tynkkynen’s Instagram post)

Meta’s January 2025 policy changes may permit some exclusionary or insulting language targeting immigrants when such speech is framed as political debate. [vii] This may explain the large number of anti-migrant comments that remained visible beneath these posts.

---

Footnotes

[i] Their were 4 posts used for this analysis all posted on the 17^th of June 2026:

abiralsahlani, Instagram, 17 June 2026, https://www.instagram.com/abiralsahlani/reel/DZr8cCREu6O/

Abir Al-Sahlani, Facebook, 17 June 2026, https://www.facebook.com/reel/1755468105445418

sebastiantynkkynen, Instagram, 17 June 2026, https://www.instagram.com/sebastiantynkkynen/reel/DZsfZt4i4Yv/

Sebastian Tynkkynen, Facebook, 17 June 2026, https://www.facebook.com/reel/2031413704463850

[ii] abiralsahlani, Instagram, 17 June 2026, https://www.instagram.com/abiralsahlani/reel/DZr8cCREu6O/

Abir Al-Sahlani, Facebook, 17 June 2026, https://www.facebook.com/reel/1755468105445418

[iii] sebastiantynkkynen, Instagram, 17 June 2026, https://www.instagram.com/sebastiantynkkynen/reel/DZsfZt4i4Yv/

Sebastian Tynkkynen, Facebook, 17 June 2026, https://www.facebook.com/reel/2031413704463850

[iv] Comments were downloaded using a third-party tool, comment replies and comments with no readable were excluded.

[v] ”Hateful Conduct”, Meta, Accessed 3 July 2026, https://transparency.meta.com/policies/community-standards/hateful-conduct/

[vi] CCDH researchers used a case-insensitive regular expression to identify comments containing variations of “send them back.” The regex matched forms of send/sent followed by up to five intervening words and then back, capturing both exact and near-phrase variants.

[vii] “More Speech and Fewer Mistakes”, Joel Kaplan, Meta, 7 January 2025, https://about.fb.com/news/2025/01/meta-more-speech-fewer-mistakes/