AI against Modern Slavery (AIMS)

Project AIMS leverages artificial (AI) techniques to help analyze corporate reporting data and promote compliance with modern slavery laws. Over time, this can lead to a viable global solution in the fight against modern slavery.

Background

Today, more than 50 million people live in circumstances involving slavery-based practices. One such practice is forced labor, a type of exploitation often embedded into corporate supply chains. Unfortunately, the complex world of supply chains is particularly opaque, which has enabled the widespread use of forced labor.

Human rights organizations use various indicators to assess the prevalence of this problem. According to the international human rights group Walk Free, G20 countries import USD$468 million worth of products annually, including electronics, garments, palm oil, solar panels, textiles and other everyday goods, considered to be at risk of being produced by forced labor.

The Modern Slavery Acts (MSAs)

The United Kingdom and Australia were among the first countries to adopt laws on modern slavery in 2015 and 2018, respectively. Canada followed suit with its own Act in 2024, The Fighting Against Forced Labour and Child Labour in Supply Chains Act. Many other countries have joined them or are considering similar legislation.

Modern slavery laws generally require large companies to publish annual reports outlining their efforts to eliminate slavery from their supply chains. This transparency helps governments and citizens hold the private sector accountable and advocate for change, both locally and globally.

The Challenge With Corporate Reporting

In the UK alone, an estimated 12,000 to 17,000 statements on modern slavery are published each year. Australia and Canada received around 3,500 and 6,000 statements, respectively, in their first reporting cycles, and these numbers are expected to rise. Without adequate resources, governments and NGOs struggle to properly review this volume of information. As a result, many statements remain unanalyzed.

As more countries adopt modern slavery laws, the number of statements submitted annually is likely to increase. However, the lack of proper analysis can significantly weaken the enforcement and impact of the law and hamper efforts to address modern slavery.

Phase 1 of the Project AIMS

The Concept

Project AIMS was initiated in 2020-2021 as part of a master's thesis by Adriana Bora, one of the 20 Rising Stars in AI Ethics and one of the Top 100 Romanians Living Abroad in the science category. A social scientist passionate about leveraging machine learning for social good, Adriana Bora began researching how AI could play a role in eliminating modern slavery.

In 2019, Adriana Bora and The Future Society partnered with Walk Free to launch phase 1 of the Project AIMS. Building on work already carried out by Walk Free, WikiRate, Business & Human Rights Resource Centre, the goal of the project was to create a methodology for analyzing statements produced by private companies under the UK Modern Slavery Act.

Using data science, machine learning techniques like natural language processing (NLP) and computational linguistics, the first phase of the Project AIMS looked for methods that would enable swift and comprehensive analysis of MSA reports. These methods could help users to obtain valuable information about slavery statements, which would be an essential step towards improving transparency within corporations and fostering continued progress in the global fight against this pressing challenge.

Facts and outcomes of phase 1

In Phase 1 of the research, the primary objective was to address challenges associated with the UK Modern Slavery Act. This phase involved determining which statements required analysis, collecting modern slavery statements, obtaining access to these documents, extracting their text, and identifying existing ground truth data. Initial data-driven experiments were conducted to mitigate limitations in the ground truth data. Additionally, a series of machine learning experiments were performed on the dataset.

Learning

Results made available open source on the Future Society’s GitHub
Learning and recommendations published at the proceedings of the AAAI Fall 2020 Symposium on AI for Social Good

Impact

Awarded the UNESCO International Research Centre in Artificial Intelligence AI Award in 2021
Findings used in the Submission to the Review of Australia’s Modern Slavery Act 2018
Project AIMS was featured as a case study for policy brief on Artificial Intelligence: Addressing or Distorting the Modern Slavery Challenge

Phase 2 of the AIMS project

Phase 2 Objectives

This second phase of Project AIMS has engaged two of the world’s most renowned AI and data science research organizations: Mila and Australia’s Queensland University of Technology (QUT)’s Centre for Data Science.

Using the Australian Modern Slavery Act as a test case, the goal of this phase was to explore how cutting-edge AI can be used to compare thousands of private sector statements.

The objectives were to

Examine the feasibility of equipping governments, NGOs and industry with a robust solution capable of analyzing large volumes of company reported data.
Gather insights that will help authorities and companies better understand where to focus resources to improve corporate compliance with the modern slavery laws.
Equip citizens and investors with the knowledge they need to make informed decisions about the companies they engage with.

The Project AIMS also hopes to inspire others to apply their expertise, whether in software engineering, data science or social sciences to the field of AI in the fight against modern slavery.

Phase 2 Outputs and Outcomes

The project has achieved substantial outcomes across research, resource development, and capacity building:

Academic Publications

Five high-quality academic papers were developed.

Two have been published in leading conferences in AI, ICLR (International Conference on Learning Representations) and ACL (Association of Computational Linguistics)
Three are currently under review

Datasets and Tools

AIMS.au: The largest annotated dataset of modern slavery statements globally, covering over 5,700 Australian modern slavery statements
AIMS.uk, and AIMS.ca: Datasets of annotated statements from the UK and Canada to evaluate whether models trained on Australian data generalise to other jurisdictions
Detailed annotation specifications documenting how statements were assessed for compliance with the Modern Slavery Act
Fine-tuned large language models achieving state-of-the-art performance in compliance assessment
A comprehensive suite of prompts (zero-shot, few-shot, chain-of-thought) for evaluating statements using generative models without fine-tuning
Benchmarks demonstrating that fine-tuned models significantly outperform prompt-based approaches and can generalise well to statements from other jurisdictions, such as the UK and Canada
Four additional teacher models were trained to mitigate specific limitations; their knowledge was distilled into a student model capable of conducting analysis approximately 7–8 times faster and with 7-fold energy savings
Explainability techniques were applied to help reviewers understand how the models arrive at their decisions

Social Science Research Contributions

A mapping of 13 unique open-sourced benchmarking methodologies of prominent studies against the mandatory criteria of the Australian Modern Slavery Act (MSA) yields over 466 metrics, capturing the diverse questions posed by academia and civil society when interrogating modern slavery statements. The intersection of these metrics forms the basis for a priority framework, grounded in literature and stakeholder interests. This framework supports assessments that move beyond basic compliance to evaluate the overall quality and substance of the statements.
An AI readiness mapping assessing the extent to which regulatory frameworks in business and human rights, those with reporting requirements, are prepared to incorporate AI methods for compliance assessment. This work also includes learnings and recommendations for the governments enforcing these laws.

Ethics and Responsible Translation to Practice

All research was designed and conducted with ethical considerations at its core, ensuring responsible development throughout. The approach emphasizes the importance of maintaining a human-in-the-loop, where final decisions are always made by human reviewers. To support this, explainable methods were prioritized to help reviewers understand and interpret the model’s outputs. This design demonstrates the potential of AI to responsibly assist human evaluators in assessing modern slavery statements across jurisdictions.
All research outputs and resources were made open source to encourage transparency, collaboration, and further innovation. To support practical uptake, the project concluded with a global hackathon featuring speakers and judges from around the world. This event invited the international research and practitioner community to build upon the openly shared resources and advance the research and develop tools that stakeholders can use to advance modern slavery reporting and analysis

Conference Papers

ICLR – AIMS.au: A Dataset for the Analysis of Modern Slavery Countermeasures in Corporate Statements
ACL Anthology – AIMSCheck: Leveraging LLMs for AI-Assisted Review of Modern Slavery Statements Across Jurisdictions

Open Resources

The Project AIMS Hackathon 2025

The Project AIMS (AI against Modern Slavery) Hackathon 2025 is a global online innovation competition that brings developers, entrepreneurs, researchers, and human rights advocates to develop AI-driven solutions to combat modern slavery. This hackathon aims to raise public awareness of modern slavery, improve participants’ technical and domain expertise, and foster interdisciplinary collaboration. The event seeks to accelerate the impact and adoption of open-source tools created through Project AIMS, a pioneering initiative that uses large language models (LLMs) to benchmark and analyse modern slavery statements issued by major corporations, thereby promoting transparency, accountability, and ethical business practices.

Part of this research was supported by the National Action Plan to Combat Modern Slavery 2020-25 Grants Program, administered by the Attorney General’s Department.

The goal of the AIMS project is twofold: empower civil society and legislators to hold businesses and governments to account; and pave the way for a new type of public policy and legislation, which embraces the full power of data mining and processing.

Nicolas Miailhe, Co-founder and President, The Future Society

Meet the Team

Adriana Eufrosina Bora

AIMS Project Lead, AI for Humanity

Adriana Bora is the Project Lead of Project AIMS at Mila, AI for Humanity and a PhD Candidate in the School of Mathematical Sciences at the Queensland University of Technology.

Adriana’s main area of expertise is in using technology against modern slavery. Her research focuses on applying machine learning in analyzing and benchmarking the businesses’ reports published following the Modern Slavery Acts from the UK and Australia.

Adriana is listed as one of the 20 Rising Stars in AI Ethics 2022 and won the UNESCO International Research Centre in Artificial Intelligence (IRCAI)’s AI Award 2021 for her efforts to develop AI solutions to combat modern slavery, contributing to SDG 8, Target 8.7.

Adriana holds a Master’s Degree in International Public Management from Sciences Po Paris School of International Affairs (PSIA). She has also studied at the University of Hong Kong and holds a diploma in International Relations and Advanced Quantitative Methods from the University of Essex in the UK.

Mila Members