Blog Archives



Submission: Research Department, Digital Rights Foundation
Submission Date: 30th November, 2023

Meta’s classifiers failed to assess the relevant context and took too abrupt a decision in removing the post shared on Instagram. The shared video showed a man confronting a woman in public because she was not wearing a hijab. The woman whose face was visible in the video, was arrested following the incident. The accompanying Persian caption used descriptive language to express the user’s support for the woman in the video and all Iranian women standing up to the regime. As per Meta's assessment, the caption was construed as having an "intent to commit high severity violence," thereby violating its Violence and Incitement policy. The post was later restored to Instagram under the Coordinating Harm and Promoting Crime policy after the user appealed to Meta and it was decided that the post did not violate community standards. This policy outlined that users may be allowed to advocate and debate the legality of content that aimed to draw attention to harmful or criminal activity as long as they did not advocate for or coordinate harm. It also outlined that any content that put unveiled women at risk would require additional information and context.

It is important to note that this context was provided in the post if the classifiers were designed to assess the content in totality instead of processing the caption and media individually. The attack did not take place in a vacuum and was a byproduct of strict moral policing by the Iranian state. This was exacerbated by ongoing political unrest that had unfolded in September 2022 when Mahsa Amini was taken into custody by the morality police under accusations of observing ‘improper hijab’ - where she suspiciously died of a heart attack. This sparked nationwide protests, united by the chant: ‘Zan, Zendegi, Azadi’ (Woman, Life, freedom).

For the ‘Woman, Life, Freedom’ movement, social media and online platforms were paramount in the mobilization of protests and broadcasting of vital information. Videos and pictures from various protests in schools, universities, and streets were circulated which showed more and more women exercising their right to freedom of expression by appearing in public without their head coverings. Social media allowed the protesters a platform to get their message out into the world. A prominent Iranian actress, Taraneh Alidoosti, posted multiple pictures of herself without a headscarf on Instagram with the caption ‘Woman. Life. Freedom’. Women willfully

unveiling in public spaces quickly became a symbol of defiance against the morality police, and the regime. As expected, acts of defiance such as unveiling in public in a political and religious climate, such as Iran’s, comes with its own risks. Women and girls who have stepped out in public without a head covering have been arrested, beaten, and had items like yogurt dumped on their heads. Men have also been arrested and beaten for showing support for the cause. 

Additionally, the Iranian authorities resorted to unprecedented levels of internet shutdowns in an attempt to silence dissent and isolate the Iranian people from the world. According to, an Iran-focused internet monitor, Iran experienced internet blackouts for over four months either nationwide, or at a provincial level after Mahsa Aminis death. Moreover, the government enacted legislation that allowed the government to monitor and identify individuals based on their online activity. 

These measures are a part of the government's effort to curtail freedom of expression and access to the global online platform. A majority of Iranian users are either experiencing constant removal of their content or know at least one person who is being censored in the Persian language. The most common type of content that has been removed or shadowbanned are hashtags of human rights campaigns; comedians posting political satire; and activists organizations using chants like “death to Khamenei”. Persian language news organizations have also had their content removed simply for discussing political organizations. Most of the content posted on Meta is in languages other than English, with more than a hundred languages being used on Facebook. This needs to be taken into account when assembling contextual embeddings. Meta needs to improve its Natural Language Processing (NLP), scaling it across more languages. Systems that detect and remove policy-violating content should then be trained accordingly.

Access to safe and well regulated social media platforms are essential for socio-political movements, making it essential for Meta to review its content moderation according to multiple regional, cultural, and linguistic contexts. Prior to content removal based solely on the judgment of automated classifiers, Meta should prioritize training manual moderators in understanding relevant complications tied to online content. This approach would allow human moderators to assess both media content and their accompanying captions in accordance to contextual cues, allowing for a more nuanced and accurate decision making process compared to evaluating them separately. Social media users ought to have the freedom to share content expressing their support for a cause or condemning harmful regimes and beliefs. This should be permitted without Meta's classifiers flagging it as a violation, even in cases where the language used may be deemed 'offensive.' It is important to note that offensive language can be used in non-offensive contexts, and hate speech does not always contain offensive language.

 CNN, Leading Iranian actor posts picture without hijab in support of anti-government protests (CNN, 2022)

 AWID, Iran's Year of Defiance and Repression: How One Woman's Death Sparked a Nationwide Uprising (AWID, 2023)

 BBC NEWS, Iranian Women arrested for not covering hair after man attacks them with yogurt (BBC, 2023)

Published by: Digital Rights Foundation in Statement

Comments are closed.