Sign in

Holli Sargeant

@hollisargeant.bsky.social
177 followers 373 following 40 posts

Research Fellow in Law & AI, St John's College, University of Cambridge Focusing on legal frameworks for algorithmic decision-making & role of AI in justice systems. hollisargeant.github.io

PostsRepliesMedia
Reposted by Holli Sargeant
Tech Policy Press @techpolicypress.bsky.social · 11/08/2026
Anna Neumann, Holli Sargeant and Jat Singh argue that system prompts alone don't predict model behavior, so AI safety assessments must evaluate a system's outputs, not solely its written instructions.
techpolicy.press
Making AI Safe Requires More Than Constraining Its Instructions
System prompts alone don't predict model behavior, so "prompt governance" is insufficient for AI safety, write Anna Neumann, Holli Sargeant and Jat Singh.
0123
Holli Sargeant @hollisargeant.bsky.social · 11/08/2026
Policymakers increasingly assume inspecting an AI’s system prompts proves it's safe. In @techpolicypress.bsky.social, we show system prompts alone don't predict model behaviour. AI safety assessments must evaluate system outputs, not solely written instructions. w/ @annaneumann.bsky.social
techpolicy.press
Making AI Safe Requires More Than Constraining Its Instructions
System prompts alone don't predict model behavior, so "prompt governance" is insufficient for AI safety, write Anna Neumann, Holli Sargeant and Jat Singh.
021
Reposted by Holli Sargeant
Tech Policy Press @techpolicypress.bsky.social · 10/08/2026
Anna Neumann, Holli Sargeant and Jat Singh argue that system prompts alone don't predict model behavior, so AI safety assessments must evaluate a system's outputs, not solely its written instructions.
buff.ly
Making AI Safe Requires More Than Constraining Its Instructions
System prompts alone don't predict model behavior, so "prompt governance" is insufficient for AI safety, write Anna Neumann, Holli Sargeant and Jat Singh.
2127
Reposted by Holli Sargeant
Pat Walshe - Privacy Matters 🇮🇪🇬🇧🇪🇺 @privacymatters.bsky.social · 28/07/2026
Reading. ‘Mind the gap: Securing algorithmic explainability for credit decisions beyond the UK GDPR’ www.sciencedirect.com/science/arti... <an important subject globally, esp in fintech imo
sciencedirect.com
125
Holli Sargeant @hollisargeant.bsky.social · 20/07/2026
Thank you @stjohnscollege.bsky.social, @cam.ac.uk for sharing my new evidence review on AI in admin, civil and family justice. The central question remains: how do we know whether AI is improving, rather than impairing, access to justice? Report 👉 doi.org/10.17863/cam... tinyurl.com/2rkfz2rx
joh.cam.ac.uk
AI is rapidly changing family and civil courts – but who checks it is right? - St John's College, Cambridge
100
Holli Sargeant @hollisargeant.bsky.social · 16/07/2026
AI is being rushed into the justice system—but is it actually helping improve outcomes? My @nuffieldfoundation.org review looks at the evidence.The short answer? We don't know. tinyurl.com/2v5a4pnb 🧵 (1/6) @adalovelaceinst.bsky.social @cam.ac.uk @cambridgelaw.bsky.social
nuffieldfoundation.org
AI rollout in UK justice system is outpacing scrutiny, new report warns - Nuffield Foundation
AI is being adopted across the justice system, despite limited evidence that it improves people’s experiences, outcomes or access to justice.
123
Reposted by Holli Sargeant
Nuffield Foundation @nuffieldfoundation.org · 14/07/2026
As AI becomes more embedded in civil and family justice and advice services, the report calls for greater transparency, stronger oversight and more independent evaluation to ensure technology supports – rather than undermines – fair and accessible justice. Find out more 👉 ow.ly/fjb350Znb8p
nuffieldfoundation.org
Artificial intelligence and justice - an evidence scoping review - Nuffield Foundation
This report examines evidence on AI and data-driven technologies in administrative, civil, and family justice settings in the UK.
012
Reposted by Holli Sargeant
Nuffield Foundation @nuffieldfoundation.org · 14/07/2026
AI has the potential to improve access to legal information and reduce administrative burdens. But the evidence base has not kept pace with its rollout. The report, from Dr @hollisargeant.bsky.social at the @cam.ac.uk, finds that very few AI tools had any publicly available evaluation.
111
Reposted by Holli Sargeant
Nuffield Foundation @nuffieldfoundation.org · 14/07/2026
🔴 AI is being adopted across the justice system faster than its impact can be fully understood. Our latest research finds that AI is being widely used across the justice system, despite limited evidence that it improves people’s experiences, outcomes or access to justice.
AI is being rapidly adopted across the UK justice system.

But there’s little evidence yet that 
it improves people’s experiences, outcomes or access to justice.
235
Holli Sargeant @hollisargeant.bsky.social · 14/07/2026
We are increasingly placing AI at the centre of plans for swifter, fairer, more accessible justice. Yet, my new report @nuffieldfoundation.org, shows that deployment is materially outpacing any independent evidence it improves access to justice. www.nuffieldfoundation.org/publications...
nuffieldfoundation.org
Artificial intelligence and justice - an evidence scoping review - Nuffield Foundation
This report examines evidence on AI and data-driven technologies in administrative, civil, and family justice settings in the UK.
012
Holli Sargeant @hollisargeant.bsky.social · 27/06/2026
@annaneumann.bsky.social just gave a fantastic presentation of our #FAccT2026 paper, "Prompt Governance? On Governing Technologies Governed by Natural Language." Read it here: doi.org/10.1145/3805...
081
Holli Sargeant @hollisargeant.bsky.social · 09/06/2026
Bumping our paper as we get closer to #FAccT2026! After a 41-day wait, our paper finally made it through the arXiv machinery! If you're interested in whether system prompts serve as robust governance, have a read and come say hi in Montréal. 🔗 arxiv.org/abs/2606.075... @annaneumann.bsky.social
arxiv.org
041
Reposted by Holli Sargeant
logan koepke @jlkoepke.bsky.social · 04/06/2026
🧵 How do companies in highly regulated domains actually test their AI systems for and mitigate discrimination? Our new @upturn.org paper at @facct.bsky.social w/ Emily Black, @mbogen.bsky.social, @s010n.bsky.social, and @wesleydeng.bsky.social, tries to answer this question: arxiv.org/abs/2606.02957
The Fair Lending Model: How the Longest-Running Algorithmic
Fairness Programs Work in Practice
EMILY BLACK∗, New York University, USA
MIRANDA BOGEN∗, Center for Democracy & Technology, USA
LOGAN KOEPKE∗, Upturn, USA
SOLON BAROCAS, Microsoft Research, USA
WESLEY DENG, Carnegie Mellon University, USA
MINGWEI HSU, Upturn, USA
U.S. financial institutions subject to fair lending laws have been running algorithmic fairness programs for decades. Despite
this long history, remarkably little is known about how these requirements operate in practice. In this paper, we offer the first
empirical account of how financial institutions test for and mitigate algorithmic discrimination on the ground. In doing so, we
shed light on how the regulatory design of fair lending law and regulation have shaped the policies, processes, and practices
of fair lending programs. Drawing on 35 semi-structured interviews with participants across the fair lending ecosystem, we
find that while financial institutions have a floor of fairness practices aimed at preventing discrimination in lending largely
absent in other domains, the specifics of how firms test for discrimination and search for less discriminatory algorithms varies
widely. We also find that regulatory supervision via fair lending examinations has been the key driver of compliance work, but
that the practical impact of fair lending programs often depends on how well they can navigate competing business incentives,
perceived legal tensions, and regulatory uncertainty. Ultimately, our findings highlight the unique role that supervisory
authority has played in successfully fostering fair lending practices—a regulatory design feature that is distinct from other
areas of civil rights law and almost completely absent from recent policy proposals for dealing with algorithmic discrimination.
CCS Concepts: • Social and professional topics → Computing / technology policy
1116
Holli Sargeant @hollisargeant.bsky.social · 03/06/2026
“Remember that all models are wrong; the practical question is how wrong they have to be to not be useful.” Usually treated as a technical question, in my new article in the Modern Law Review, I argue it is also a legal one. doi.org/10.1111/1468...
2155
Holli Sargeant @hollisargeant.bsky.social · 21/05/2026
Can you regulate AI with its instructions? Our #FAccT2026 paper (w/ @annaneumann.bsky.social & Jat Singh) tests that assumption against research literature, EO 14319 & EU GPAI Code.
140
Reposted by Holli Sargeant
Maria Antoniak @mariaa.bsky.social · 28/01/2026
"We analyze all papers published at ACL, NAACL, and EMNLP in 2024 and 2025... nearly 300 papers contain at least one HalluCitation... Notably, half of these papers were identified at EMNLP 2025 ... indicating that this issue is rapidly increasing." www.arxiv.org/abs/2601.18724
arxiv.org
HalluCitation Matters: Revealing the Impact of Hallucinated References with 300 Hallucinated Papers in ACL Conferences
Recently, we have often observed hallucinated citations or references that do not correspond to any existing work in papers under review, preprints, or published papers. Such hallucinated citations pose a serious concern to scientific reliability. When they appear in accepted papers, they may also negatively affect the credibility of conferences. In this study, we refer to hallucinated citations as &#34;HalluCitation&#34; and systematically investigate their prevalence and impact. We analyze all papers published at ACL, NAACL, and EMNLP in 2024 and 2025, including main conference, Findings, and workshop papers. Our analysis reveals that nearly 300 papers contain at least one HalluCitation, most of which were published in 2025. Notably, half of these papers were identified at EMNLP 2025, the most recent conference, indicating that this issue is rapidly increasing. Moreover, more than 100 such papers were accepted as main conference and Findings papers at EMNLP 2025, affecting the credib
34013
Holli Sargeant @hollisargeant.bsky.social · 20/01/2026
The UK has repealed Article 22 of the GDPR. What replaces it? For Oxford Business Law Blog, I argue the Data (Use and Access) Act 2025 creates 3 structural gaps that leave consumers vulnerable to automated decisions. Blog: shorturl.at/HRH5f Article: doi.org/10.1016/j.cl... #LawSky #AIPolicy
shorturl.at
Mind the Gap: Securing Algorithmic Explainability for Credit
184
Holli Sargeant @hollisargeant.bsky.social · 05/12/2025
Automated decisions are increasingly made by opaque models—the UK’s DUAA amendments to #GDPR makes them even harder to challenge. My new article in @elsevierconnect.bsky.social CL&SR sets out what’s at stake and how to fix the explainability gap. #LawSky doi.org/10.1016/j.cl...
doi.org
051
Holli Sargeant @hollisargeant.bsky.social · 04/12/2025
2025 has been a turning-point year for the “right to explanation” and GDPR Article 22. From C-203/22 Dun & Bradstreet, to the UK’s new Data (Use and Access) Act 2025, to the EU’s Digital Omnibus proposals — automated decision-making rules are shifting fast. 🧵 1/4
100
Holli Sargeant @hollisargeant.bsky.social · 12/11/2025
Honored that our paper “Detecting Legal Citations in United Kingdom Court Judgments” was featured in the Senior Area Chair Highlights at #EMNLP2025! Thanks to my co-authors Andreas Östling and @mansmag.bsky.social and to the reviewers and SACs for the recognition. aclanthology.org/2025.emnlp-m...
aclanthology.org
Detecting Legal Citations in United Kingdom Court Judgments
Holli Sargeant, Andreas Östling, Måns Magnusson. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.
031
Holli Sargeant @hollisargeant.bsky.social · 04/11/2025
Our #EMNLP2025 paper, “Detecting Legal Citations in United Kingdom Court Judgments,” presents the first systematic study of legal citation detection for UK case law and legislation. aclanthology.org/2025.emnlp-m... @mansmag.bsky.social
aclanthology.org
Detecting Legal Citations in United Kingdom Court Judgments
Holli Sargeant, Andreas Östling, Måns Magnusson. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.
232
Reposted by Holli Sargeant
Law, Ethics & Policy of AI - Blog & Summer School @kul-ai-sschool.bsky.social · 14/10/2025
The Law, Ethics & Policy of AI Blog is back! In a new post, Arina Shah @hollisargeant.bsky.social @mackenziej.bsky.social Adrian Weller&Umang Bhatt explain how #uncertainty in #AI can lead to discrimination and why selective friction may better than abstention 🔗 www.law.kuleuven.be/ai-summer-sc...
073
Reposted by Holli Sargeant
SSRN @ssrn.bsky.social · 06/08/2025
Check out the latest from the SSRN #blog which includes a selection of #research on #Cyberspace Law. Read More: spkl.io/63321AJgCd #Academicsky #Lawsky
spkl.io
The Latest Research on Cyberspace Law
This list includes a selection of the latest research on Cyberspace Law posted to SSRN in 2025. Seeing is Believing? Deepfakes in Financial Markets by Michal Lavi (Hebrew University of Jerusalem) &…
133
Holli Sargeant @hollisargeant.bsky.social · 23/06/2025
I never need an excuse for summer in Greece, but I'm happy to be in Athens for #FAccT2025, where I'll be presenting two exciting new papers this week! Can't wait to connect with everyone in Athens!
110
Holli Sargeant @hollisargeant.bsky.social · 10/06/2025
📢 New Paper! Check out my paper "From Estimation to Discrimination: Algorithmic Bias, Predictive Uncertainty, and Anti-Discrimination Law" that explores how different types of predictive uncertainty can cause unlawful discrimination. dx.doi.org/10.2139/ssrn... #AcademicSky #LawSky
dx.doi.org
From Estimation to Discrimination: Algorithmic Bias, Predictive Uncertainty, and Anti-Discrimination Law
Machine learning (ML) systems, increasingly deployed in high-stakes decisionmaking, inherently produce uncertain outputs that can lead to unlawful discriminatio
060
Reposted by Holli Sargeant
Jennifer Cobbe @jennifercobbe.bsky.social · 03/06/2025
Amazing opportunities to work with Jat Singh on responsible AI, for both a PhD student and a postdoc - he's particularly keen to find someone from a law background interested in FAccT-related themes More info - PhD: www.compacctsys.net/RespAI-PhD.pdf Postdoc: www.compacctsys.net/RespAI-Postdoc.pdf
compacctsys.net
177
Holli Sargeant @hollisargeant.bsky.social · 14/04/2025
So excited to have two papers at @facct.bsky.social w @mansmag.bsky.social! 📄 One paper formalises UK anti-discrimination law in framework for automated decision-making: arxiv.org/abs/2407.00400 💡 The other explores transformers for better hate crime classification in Sweden. See you in Athens!
arxiv.org
Formalising Anti-Discrimination Law in Automated Decision Systems
Algorithmic discrimination is a critical concern as machine learning models are used in high-stakes decision-making in legally protected contexts. Although substantial research on algorithmic bias and...
051
Holli Sargeant @hollisargeant.bsky.social · 26/02/2025
🚀 Just Published! Check out our new article in Artificial Intelligence and Law where we use Claude 3 Opus to classify and analyse UK summary judgment cases—introducing a new functional taxonomy tailored for UK legal topics. 🔗 doi.org/10.1007/s105...
doi.org
Topic classification of case law using a large language model and a new taxonomy for UK law: AI insights into summary judgment - Artificial Intelligence and Law
This paper addresses a critical gap in legal analytics by developing and applying a novel taxonomy for topic classification of summary judgment cases in the United Kingdom. Using a curated dataset of ...
020