A Berkeley Human Rights Center study argues that large language models can affect rights across law, journalism and education, and calls on developers and deployers to assess those effects. Its researchers interviewed 56 people in 24 countries, bringing evidence from professions and communities often left out of the model-building conversation.
Watch Desk analysis
What happened
The study, described by Quantum Zeitgeist, examined both benefits and risks of generative AI in law, journalism and education. The researchers say LLMs can speed up work, but can also let people skip research, organisation and critical thinking. They singled out the effect on students learning foundational skills such as reading, research, outlining and revision.
Their recommendations include human-rights impact assessments by developers and deployers, tailored to each group’s role. The researchers also call for assessment across the development and deployment process, from upstream data collection and labelling to monitoring systems in use.
Key findings
- AI can save time, but shortcuts have a cost
Researchers cited journalists using LLMs to produce sports reports quickly, while warning that bypassing research and analysis can threaten accuracy and thoroughness. - Students may miss essential practice
The researchers highlighted risks to young people learning reading, research, outlining and revision, skills that support independent thought and comprehension. - Responsibility depends on the role
The report calls for developers and deployers to assess human-rights impacts in ways suited to their different responsibilities. - Risks and benefits vary by context
Interviewees’ views differed across regions, underlining that a single assessment may miss how impacts vary between communities.
Why it matters
This is a practical attempt to turn human-rights principles into questions AI teams can ask before and during deployment. The researchers say assessments should consider factors such as the scope, scale and likelihood of potential impacts, rather than treating rights as an abstract concern to be revisited after a system is already in use.
The work is based on interviews, not a measurement of how common each harm is across all users or settings. Still, its central point is useful: efficiency is not the only result that matters, especially in fields where people’s rights, learning and access to reliable information are at stake.
Our read
The strongest recommendation is also the least glamorous: work out who could be affected, at what stage, and who is responsible for responding. That is harder than adding a principle to a policy page, but rather more useful. Developers and organisations deploying LLMs should treat rights assessments as an ongoing part of the work, not a one-off sign-off.
What to watch
- Whether developers and deployers adopt the report’s role-specific assessment approach.
- How teams translate scope, scale and likelihood into practical checks.
- Whether the planned webinars lead to concrete uptake beyond the research community.
- How future evidence distinguishes local benefits from harms, particularly for students and professional users.
Discussion spark: Should organisations have to publish human-rights assessments before deploying LLMs in schools, newsrooms and legal services, or would mandatory disclosure add paperwork without meaningful accountability?
Sources and evidence
- Berkeley Law Report Finds AI Use Affects Rights In Law, Media, Schools – Quantum Zeitgeist (29 September 2026, 08:24 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.