An AI-Based Approach to Early Reporting and Justice Initiation in Image-based Sexual Abuse. A Pilot Study

In the digital age, the internet has become a double-edged sword, offering connection while simultaneously creating new avenues for profound harm. One of the most distressing forms of this harm is image-based sexual abuse, a crime where intimate or private images of a person are shared online without their consent. This can happen through the hacking of personal devices, the non-consensual distribution of photos shared in trusted relationships, or the creation of realistic but fake images using artificial intelligence. For the survivors of these acts, the trauma is often compounded by the feeling of being exposed to the entire world, leading many to withdraw and avoid seeking help. The path to justice is frequently blocked by a specific, paralyzing hurdle: the difficulty of writing a formal report. Survivors often face what experts call "blank page anxiety," a state of distress where the fear of describing the event perfectly, in the correct legal language, prevents them from taking the first step to report the crime to police or online platforms.

Researchers Mattia Falduti and Anca Radu have explored a way to remove this barrier using a new kind of digital helper. Their work focuses on a pilot study of an artificial intelligence tool designed specifically to assist survivors of image-based sexual abuse in drafting their initial reports. The core idea is simple yet transformative: if a computer can ask the right questions and turn a survivor's story into a formal document, it might be possible to get more people to seek justice. The researchers built a digital agent that acts as a supportive interviewer, guiding the user through a calm conversation to gather the essential facts needed for a police report or a request to remove content from a website. This tool does not replace human lawyers or police officers; instead, it serves as a bridge, translating a survivor's raw experience into a structured text that authorities can understand and act upon.

The team designed this tool to be accessible and safe, operating as a no-code online solution that uses advanced language processing to communicate in multiple languages. The process begins when a user arrives at the tool and selects their preferred language. The digital agent introduces itself clearly as a support resource for survivors and then begins a gentle, step-by-step inquiry. It asks for specific details such as where the survivor is currently located, where the incident took place, and when the abuse was discovered. It also asks for a brief description of what happened and whether the survivor has a link to the abusive image. Throughout this interaction, the agent is programmed to avoid asking for sensitive personal identifiers like names or addresses, ensuring the user feels safe. Instead, it collects the factual backbone of the story, leaving a space for the user to add their personal details later when they are ready to submit the report to the authorities.

Once the necessary information is gathered, the tool performs its most critical function: it synthesizes the answers into a formal draft report. This document is structured to meet legal standards, containing all the temporal and spatial details required to initiate an investigation or request the removal of content. The agent then presents this draft to the user, summarizing the facts to ensure accuracy before asking for confirmation. The final output is a text that the survivor can copy and paste directly to law enforcement agencies or digital platforms. The tool also provides immediate, practical guidance, reminding the user to fill in their personal information and providing emergency contact numbers, such as 112 for Europe or 911 for the United States, to ensure they know exactly how to proceed.

To test whether this approach actually works, the researchers invited three legal experts to interact with the tool and evaluate its output. These experts included a lawyer with court experience, a law enforcement officer with thirty years of service, and a legal professional with a background in both the courts and the public prosecutor's office. The experts were asked to judge whether the information collected was sufficient to start a legal process, if the instructions were clear, and if the generated text could be used to report the abuse. The results were overwhelmingly positive. All three experts agreed that the information gathered was enough to begin the justice process. They found the instructions accurate and the generated reports usable for contacting authorities. One expert noted that the tool could effectively "set the wheels of justice in motion," while another highlighted that the minimal time required to create the draft was a significant advantage for survivors who are often overwhelmed.

Despite the promising results, the researchers are careful to frame this as a pilot study rather than a final solution. The evaluation was conducted with a small group of experts, and the authors acknowledge that the tool needs to be tested with actual survivors to see how it performs in real-world scenarios. They also suggest that future versions could include more comprehensive support, such as links to helplines and legal aid services, and that the tool might eventually be embedded directly into the websites of digital platforms where the abuse occurs. The study suggests that current artificial intelligence solutions have the potential to guide survivors toward the right path immediately after an incident, transforming a digital environment that often feels like a source of trauma into a space that offers real, effective, and accessible help. By addressing the initial hurdle of writing a report, this approach offers a tangible way to facilitate access to justice for those who have suffered image-based sexual abuse.