AI creates risk, opportunities for key human rights fields, report finds

A new report from the Human Rights Center at UC Berkeley School of Law evaluates how large language models (LLMs) influence human rights across law, journalism, and education. The assessment, based on interviews with 56 experts from 24 countries, highlights how LLM use varies globally, with practitioners in the Global North often more risk-averse than those in the Global South. Researchers identified both risks and opportunities tied to geographic and socioeconomic contexts, emphasizing the need for context-specific approaches to LLM deployment.

The report warns that overreliance on LLMs can undermine critical thinking and exacerbate cultural homogenization, posing threats to freedom of thought and equality. In law, LLMs may assist unrepresented litigants but risk compromising due process if misused in high-stakes settings. Journalism could benefit from improved productivity, yet hallucinations risk spreading misinformation. Education may see gains in personalized learning, though alignment with standards and recall support remain inconsistent.

Researchers recommend diversifying training datasets to better reflect global communities and designing LLMs with refusal mechanisms tailored to human rights risks. They also urge the development of profession-specific models and clear remediation processes to address harms. The report calls for comprehensive risk assessments throughout the AI lifecycle and targeted regulations for high-risk applications, such as judicial use of LLMs.

The findings build on prior work, including a framework for evaluating human rights risks in AI models presented at the AAAI/ACM Conference on AI, Ethics, and Society (AIES) 2025. Co-authors also explored the global implications of LLM development in a separate publication, underscoring the urgency of integrating human rights considerations into AI governance.