Sexual Violence Awareness-Raising Posts Should Not Be Buried Just Because They Are ‘Sensitive.’ Meta’s Rules Need Changing
20 अगस्त 2026
By Nighat Dad, Oversight Board Member
Survivors of sexual violence and their advocates rely on Meta’s platforms to share their experiences and raise awareness about women’s safety. Yet Meta’s policies and the way they are applied are automatically reducing the reach of this content. In doing so, the company is ignoring how survivors, advocates and their intended audiences are using its platforms to fight sexual violence.
The Board’s latest decision, Posts Condemning Bus Harassment, looked at two instances where Meta reviewers removed videos, one on Facebook of a woman being groped by a man on a bus in India and another on Instagram of a man in Brazil masturbating on a bus. In each case, a woman sitting next to the man filmed the incident. After the Board started working on the appeals against the rulings, Meta restored the videos, each with a “Mark as Sensitive” warning screen and age-gating.
Reduced Reach
Correcting that enforcement error should be the end of the matter. But it isn’t. As Meta told the Board, when given the warning screen, content automatically becomes non-recommendable so its reach can be sharply reduced. That is despite the company eventually agreeing that, in the Indian case, the post had been made for awareness-raising reasons.
India's 2012 Nirbhaya incident of a gang rape on a bus drew worldwide attention. Brazilians will be acutely aware of the 2017 case of public masturbation on a bus there that went viral and sparked outrage. Amid statistics showing high rates of violence against women in those two countries, there is heightened concern for their safety in public spaces. The lead United Nations body on gender equality, UN Women, reports that nearly 90% of women and girls experience some form of violence on public transportation. Social media plays a critical role in education and advocacy on such social issues and safety concerns.
The automatic non-recommendation status, paired with a warning screen, strips away safeguards that apply everywhere else in the system.
The automatic non-recommendation status, paired with a warning screen, strips away safeguards that apply everywhere else in the system. If content is taken down, Meta tells users and provides a reason and means to appeal. However, prohibiting something from being recommended is different. Users cannot appeal it. It is not even clear if they are made aware of the non-recommendable outcome of their content. The Board said coupling the “Mark as Sensitive” warning screen with the non-recommendable status is an “unnecessary additional restriction on freedom of expression.” It has said this in past cases too.
Four rounds of human review, including subject-matter experts, assessed the Indian content; two rounds assessed the Brazilian content. Then, after the Board intervened and Meta overruled the reviewers’ decisions, automation limited how far either post could travel anyway.
The layers of process kept the content from being seen, in two ways. Content documenting sexual violence sits, by necessity, close to what Meta polices as adult sexual exploitation, nudity or sexual activity. Ambiguous cases default toward restriction unless there is a specific exception. The Indian case had that exception for this harm, but it wasn’t applied. The Brazilian case had no exception available but there should have been one.
Right Rules Needed
The right rules need to be in place so that reduced reach of awareness-raising content doesn’t happen in the first place. More reviews, AI or human, stacked on top of the assessments already made will not help. The rules were unclear, leading to confusion. Under Meta’s Adult Sexual Exploitation policy, there is an exception allowing depictions of non-consensual sexual touching, if it is not in a “sensational context.” But “sensational context” is undefined, for moderators and users.
In the Brazilian case, the content was removed because Meta’s internal guidance says any content depicting masturbation in public that mocks, sexualizes or identifies an individual, including content intended to condemn the behavior, violates its Adult Sexual Exploitation policy and must be removed (despite people considering public masturbation to be sexual harassment and intimidation). But at the same time, the public-facing policy recognizes the importance for “survivors to share their experiences.” These cases required independent oversight, outside the enforcement process, to assess whether the rules are correct, diagnose the issues and present the solutions.
These cases required independent oversight, outside the enforcement process, to assess whether the rules are correct, diagnose the issues and present the solutions.
Independent oversight is also needed to scrutinize the AI systems that rank and recommend content at scale. Such a ranking system decided, in the Indian case, how far a public interest post on sexual violence in a public space from a Facebook page of a prominent news organization with 1.1 million followers would travel.
The Board’s decision underscored that these systems are rarely able to distinguish between engagement driven by concern, empathy or public interest and engagement driven by outrage or other motivations. That means genuine awareness-raising content can be buried due to being “sensitive” and, at the same time, no distinction is made for entities that exploit content purporting to be awareness-raising to engagement-bait and drive revenue.
“Mark as Sensitive” warning screens should not automatically make content non-recommendable. Users need clarity over which ways they are permitted to raise awareness about significant issues like sexual violence and women’s safety. Moderation systems now decide which speech reaches an audience. They carry none of the notice, reasons and appeal obligations that attach to removal. That gap needs closing now, across social media and AI platforms alike, before it is built into the next generation of these systems.
