The first time I went to interview the mother of a disappeared man in Lebanon, she did not begin by telling me how her son had been taken. She showed me his room. The bed was made. His clothes were clean and neatly folded. She had washed them again and again over the years, keeping them ready for the day he might walk through the door.
Her son had disappeared during Lebanon’s civil war, which ended in 1990. When I met her in 2014, decades had passed. She was still waiting. I remember leaving her home wondering how someone grows old without knowing whether their child is alive or dead, and how you begin to mourn when there is still a chance, however small, that they might return.
At the time, I was working with Act for the Disappeared, a Lebanese human rights organization collecting families’ stories for Fushat Amal (‘Space for Hope’ in English), a digital memorial dedicated to people who went missing during the war. Behind the often-cited estimate of 17,000 missing people were individuals with names, faces and families, whose lives deserved to be remembered.
Years later, that encounter was one of the reasons I began exploring artificial intelligence (AI) and transitional justice. I wanted to understand what the future of this field might look like. Could new technologies help that woman find out what happened to her son? Could they offer anything to the thousands of other families still searching in Lebanon?
I did not expect technology to solve what politics had failed to address. Lebanon’s missing persons issue remains deeply entangled with the legacy of the civil war and the reluctance of those who may hold information to disclose it. But I wanted to know whether new investigative possibilities were emerging, and what they might mean for people still waiting for answers.
Machine learning and satellite imagery
Over the past three years, I have spoken with organisations exploring and using AI in investigations, documentation and other justice efforts. I have come across projects that show how these technologies can help researchers and investigators work with large amounts of information, identify patterns and pursue leads.
In Mexico, researchers have used machine learning and satellite imagery to identify areas that may contain clandestine graves. In Uruguay, the university-led Cruzar project is using AI to help analyse dictatorship-era archives, working alongside relatives of the detained-disappeared.
Who is this actually helping?
AI is not one single tool, and how it works depends on what it is being used for and the information available. Take the example of analysing historical archives. Imagine thousands of pages of documents that would take people years to read and compare. Depending on the tool, AI can help turn scanned pages into searchable text, identify names, places and events, and suggest connections between records that people can then examine.
But the results are not necessarily correct. AI may miss important context, make mistakes or produce a convincing answer that is wrong. With some AI systems, it can also be difficult to understand how a particular result was reached, even for the people who developed them. That makes checking the results, and understanding the limitations of the tool being used, essential.
These examples interested me because they show how AI might help address some of the challenges involved in uncovering the truth about past violations. But identifying a possible burial site is not the same as establishing who is buried there. Finding connections in an archive does not replace the work of verifying evidence or reconstructing what happened.
I kept coming back to a question that is easy to overlook when discussing technological innovation: who is this actually helping?
The risks are not abstract
When I was collecting stories for Fushat Amal in 2014 and 2015, we were not thinking about how those photographs and testimonies might one day be analysed using AI. Families shared them so their loved ones would be remembered. Would they want that information used in an AI-assisted search today? We never thought to ask.
Some families might welcome new ways of searching; others might worry about where their information could end up or how it might be used.
At Impunity Watch, we worked in 2025 on a mapping report examining more than 40 cases of AI and data-driven technologies in transitional justice. Alongside promising applications, the report identified concerns about bias, the protection of sensitive information, inadequate safeguards and the exclusion of victims and survivors from decisions about technologies that could affect them.
To understand what these risks might mean in practice, imagine a human rights organisation working in a country emerging from conflict. It has collected hundreds of testimonies from survivors of a massacre and wants to use AI to organise the documents, identify names and places, and find connections between different accounts. This could help the organisation make sense of information that might otherwise take years to analyse.But what if a person’s name appears differently across testimonies because it has been transliterated from another language? The AI system might fail to recognise that the records refer to the same person, or mistakenly connect two different people. What if an AI-generated summary hallucinates and includes a detail no witness ever mentioned? If that information is not checked and finds its way into an investigation or public report, it could distort the account of what happened.
Who is responsible if someone is harmed?
There are also risks for the people who gave their testimonies.
Imagine that an AI tool removes a witness’s name from a document but leaves details about their home, family or circumstances that make them identifiable. Or that sensitive information is leaked or accessed by people who want to silence witnesses. In a country where those responsible for past violence may still hold power, the consequences could include surveillance, threats or retaliation.
Who would be responsible if someone were harmed? The organisation that collected the testimony, the company that provided the AI tool, or those who failed to protect the information? And what could the person affected do to seek redress? These are people who may already have suffered serious violations. Efforts to document those violations should not put them at risk of further harm.
What makes these questions particularly important in transitional justice is the context in which these technologies are being used. The information may concern serious human rights violations, the people sharing it may still be at risk, and trust in institutions may already be fragile. Yet documentation groups, victims’ associations and local archives may lack the legal or technical resources to assess every new tool. A technology that is useful in one context may be inappropriate in another.
When AI is involved in violations
There is another question that transitional justice may increasingly have to confront: not how AI is used to investigate past violations, but how it may have been involved in those violations in the first place. AI is being used in surveillance and military targeting. If someone is killed following an incorrect AI-assisted identification, how can future investigations reconstruct the decisions that led to their death, and who should be held accountable? AI does not remove human responsibility, but it can make those decisions harder to examine.
That is why decisions about AI in transitional justice cannot be left to technology developers alone. They require collaboration between people with expertise in technology, human rights, law, documentation, ethics and psychology and the contexts in which these tools will be used. Victims and affected communities also need to be involved from the beginning, when deciding what problem needs to be addressed and whether AI is the right approach.
A need for safeguards and regulation
The need for safeguards suited to these sensitive contexts is emphasised in the July 2026 report by UN Special Rapporteur Bernard Duhaime on emerging technologies and transitional justice. He calls for stronger regulation of technology companies and specific guidance for the use of emerging technologies in these contexts, including human rights impact assessments, data protection and professional standards for practitioners. The importance of such safeguards is also reflected in Impunity Watch’s broader work on AI and transitional justice.
That guidance is needed. But organisations do not have to wait for it to ask some basic questions: What did people originally agree to when they shared their stories? Is AI necessary for the work being proposed? What could go wrong, and what safeguards are in place to protect them?
I do not know whether the mother I met ever learned what happened to her son. I hope she did. I still remember his room and the clothes she kept ready. When I began researching AI, I wanted to know whether the future might offer something that had not been possible when I met her.
Three years later, I have seen approaches that I would not have imagined then, and questions that deserve much more attention. I cannot say whether any of these tools could have helped her. But I still want to find out what they might make possible for families who are searching today.
Michelle Bouchebel is the Thematic Lead on AI and Emerging Technologies at Impunity Watch, leading work on how new technologies are reshaping transitional justice. Before joining Impunity Watch, she worked with Search for Common Ground in Lebanon, leading teams implementing programmes across Lebanon and Syria on gender, governance, cultural protection and conflict transformation. She has also worked with Act for the Disappeared in Lebanon and the International Center for Transitional Justice.
