Over the Summer, I worked with Professor Tracey Holland on her project assessing the state of children’s rights and human rights due diligence (HRDD) in the arms industry.
In recent years, across many industries, companies have adopted voluntary human rights policies intended to protect their employees, and the communities that they affect, in line with international frameworks like the United Nations Guiding Principles on Business and Human Rights (UNGPs). The defense industry however has historically been reluctant to enact effective HRDD policies, particularly with regards to the use of their products by state actors, arguing that it is the responsibility of state regulators to protect human rights by imposing export controls on weapons. Increasingly this view has begun to shift, as some defense companies have acknowledged the reputational, and potentially legal risks, posed by their potential complicity in human rights violations. But at the moment the defense sector takes an inconsistent and varied approach to human rights in their corporate policies.
The part of the project I have been working on has primarily involved analyzing prominent defense firms to create a ranked index assessing firms based on their compliance with HRDD standards, and to identify where the industry is currently falling short, with the ultimate goal of providing a helpful tool for investors. At the start of the Summer, I worked with another student researcher (Lydia Mhando, 2029), building off of research done by previous years’ Ford Scholars, to assess a set of companies by hand. We read companies’ publicly released documents, like annual shareholder reports, codes of conduct, and official human rights policies, to gather evidence and qualitatively assess the strength and extent of their policy commitment and system for mitigating risk, monitoring end-use, and carrying out investigation and remediation in cases of human rights abuse.
I then built on the work of a previous researcher, Rishika Mamidibathula (2027), who had built a program that uses Claude to pull all available information about companies’ policies online and score them according to set criteria based on UNICEF/UNGPS/other international frameworks. The program also produces separate “reasoning” and “audit” tabs explaining its decision-making and its sources. I discovered that the program was not actually fetching data from the web, but was fabricating company policies based on its training data. Additionally, it was not limited to just analyzing company documents, but would have analyzed news stories and reports by NGOs about companies, somewhat different from the data this part of the project is trying to focus on. I tried to implement a web fetch component, as well as a system to check the fidelity of direct quotes that the program sourced from company documents, and I limited the program to gather evidence from each company’s respective domains. I also attempted to limit the time and cost by using batch processing. I was then able to run the program for the top 100 defense firms globally.
So far, I have found that the program has been useful for finding a huge amount of evidence on company documents, and is helpful for showcasing some wider trends in the industry, but the fidelity check process has not worked, and on reviewing the output, I have found that it sometimes misses key corporate documents.

This summer has been a great opportunity for me to gain experience doing research on a fascinating topic. There are still several areas that I would like to pursue in the future, particularly with regards to improving the program, as well as reaching out and having conversations with people who work in the defense industry to understand how their companies have developed HRDD policies.



























