The Platform Question: Power, Accountability and Global South Andean region and the Southern Cone were left adrift. There was no longer anyone human to refer cases involving their platforms to. Eva Cruells, founder and coordinator of Fembloc, a feminist initiative dedicated to combating digital genderbased violence, told the Spanish media outlet Público: “The thing is, the reporting mechanisms are very opaque and often unresponsive. We have our own data: of the total number of cases we have reported, only 4% have received any affirmative response from the platform. Because often they are bots or automated emails. Meta, for example, has greatly reduced its content moderation staff.” We revealed a similar situation as NGO Amaranta in 2020, with the study Chile and gender-based violence on the internet,[4] where we revealed that 11% of victims of doxing, digital harassment, cyberflashing, defamation, and loss of access to their account or unauthorized access to it, tried to report it to the platforms without success. When victims were asked what actions they took with the platforms, the vast majority said that the reporting sites were in English, did not respond to them, or had bugs that made them return to the same page to fill in their details over and over again.[5] Who takes care of what we don’t want to see on the internet? The history of digital moderation reveals not only the technical complexity of the process, but also its profound human inequalities. 34 The history of digital moderation reveals not only the technical complexity of the process, but also its profound human inequalities. Meta began filtering content in 2007, after New York Attorney General Andrew Cuomo demanded that sexual images proliferating on Facebook be removed within 24 hours. Since then, the company has outsourced the task to subcontractors, some of which, such as oDesk, paid moderators in Morocco as little as $1 an hour, a situation that is repeated in several countries. Behind the algorithms that protect the audience, there are people exposed daily to traumatic content without psychological support or decent working conditions. A 2019 article in The Verge reveals how workers at a Cognizant subsidiary in Tampa faced precarious working conditions, nine-minute breaks, the death of a coworker from a heart attack at his desk, and even how they had to explicitly approve certain violent content simply because it did not comply with Facebook’s moderation policies.[6] Many A Big Tech Accountability in Content Moderation with Feminist Perspective The Platform Question: Power, Accountability and Global South of these workers joined the platform under the promise of job stability and believing that they were performing a protective role for social media users. However, one animal rights activist worker recounts in the article that he had to watch a video showing a case of extreme animal torture by teenagers repeatedly. The platform’s policies did not allow him to take it down, and his superiors consoled him by pointing out that keeping it posted could prompt law enforcement authorities to launch an investigation and find the culprits. That never happened. The case highlights how moderation policies, rather than protecting, end up reproducing other forms of violence and trauma. For example, it is not possible to remove a sexually aggressive video without a valid argument as to why it cannot be shown. Spanish lawyer Francesc Feliu argues that this precision is intended to train AI. “It cannot be trained if the content it feeds on is violent, pedophilic, or murderous,” he adds. Feliu legally represents more than forty workers dedicated to moderating META content in Spain, who have jointly sought compensation for the physical and mental damage to which they are exposed in this work. The lawyer points out that most of his clients suffer from anxiety disorders, depression, and post-traumatic stress, and that several are even admitted to psychiatric centers. All of them live with the pressure of not being able to talk openly about the problem, out of shame or fear of reprisals from the company, due to the confidentiality clauses they signed.[7] This reality is repeated on other platforms such as TikTok. In 2022, it was revealed that some of the staff responsible for moderating content must follow accelerated review protocols (sometimes even seconds per video) to keep up with the massive volumes of posts they are asked to analyze every day[8] in categories such as “explicit violence,” “hate speech,” and “sensitive content.”[9] Automated language models are no exception to this reality. The work of content moderators for ChatGPT follows this same pattern of bad practices and job insecurity. In Kenya, workers have revealed that they review up to 700 passages of text per day, many of them laden with explicit sexual violence, suicide, murder, child abuse, and bestiality. They report having lost their families in the process and not having any psychological support to cope with the situation. While large companies advertise technological advances that promise to free society from arduous tasks, that same development depends on people who bear the emotional burden of working with the dark side of the internet. While large companies advertise technological advances that promise to free society from arduous tasks, that same development depends on people who bear the emotional burden of working with the dark side of the internet. A Big Tech Accountability in Content Moderation with Feminist Perspective 35