Madelyn Rose Sanfilippo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Madelyn Rose Sanfilippo has a total of 55 co-authors. Here are the top co-authors by number of shared publications:
- Yan Shvartzshnaider
Department of Electrical Engineering and Computer Science, Lassonde School of Engineering, York University, North York, ON, Canada
7 shared publications - Kyrie Zhixuan Zhou
University of Texas at San Antonio
5 shared publications Zachary Kilhoffer
3 shared publications- Abhinav Choudhry
University of Illinois at Urbana-Champaign
2 shared publications Brett M. Frischmann
Villanova University, Pennsylvania
2 shared publications- Ece Gumusel
Indiana University Bloomington, USA
2 shared publications
Latest Publications
- Open Access
- 21 October 2026
- Published by Elsevier in SSRN Electronic Journal
Sports Blackouts in the Streaming Era
Technological change has played an important role in the history of professional sports, including around player's rights, broadcasting, and fan access. This article draws on digital humanities, information policy, and everyday information behavior scholarship to shine a light on how these patterns have led to consolidation of power and money by leagues and team owners, yet streaming offers interesting opportunities for player autonomy and choice, as well as fans more power and choice in accessing game broadcasts.
- 21 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSports Blackouts in the Streaming Era
Technological change has played an important role in the history of professional sports, including around player's rights, broadcasting, and fan access. This article draws on digital humanities, information policy, and everyday information behavior scholarship to shine a light on how these patterns have led to consolidation of power and money by leagues and team owners, yet streaming offers interesting opportunities for player autonomy and choice, as well as fans more power and choice in accessing game broadcasts.
- Open Access
- 30 September 2026
- Published by Elsevier in SSRN Electronic Journal
Polycentric Governance Gaps in U.S. AI and Privacy Law
Paper presented at 2026 TPRC | The rapid proliferation of generative AI is fundamentally reshaping the U.S. legal and regulatory landscape, introducing significant uncertainty for developers and policymakers navigating the complex interplay of federal directives and state-level initiatives. Despite the growing body of literature on AI's technical vulnerabilities, there remains a critical gap in understanding the compliance environment and the efficacy of current state-level regulatory responses. This paper employs a qualitative legal and policy analysis of 200 enacted state bills collected between October and December 2025 to examine: areas of emerging state consensus on AI privacy; the consequences of polycentric compliance; and existing governance gaps. Our analysis reveals four persistent gaps in the U.S. regulatory architecture: a lack of explicit oversight regarding inferential analytics; limited remedies for algorithmic bias; the exclusion of emerging risks (e.g., deepfakes and synthetic media); and highly heterogeneous enforcement mechanisms that rely on varying degrees of regulatory and private oversight. These governance gaps contribute to a fragmented compliance environment that disproportionately impacts vulnerable populations and undermines accountability. This study ultimately highlights the urgent need for harmonized frameworks that transcend the limitations of current polycentric approaches to ensure equitable and accountable AI governance.
- 30 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPolycentric Governance Gaps in U.S. AI and Privacy Law
Paper presented at 2026 TPRC | The rapid proliferation of generative AI is fundamentally reshaping the U.S. legal and regulatory landscape, introducing significant uncertainty for developers and policymakers navigating the complex interplay of federal directives and state-level initiatives. Despite the growing body of literature on AI's technical vulnerabilities, there remains a critical gap in understanding the compliance environment and the efficacy of current state-level regulatory responses. This paper employs a qualitative legal and policy analysis of 200 enacted state bills collected between October and December 2025 to examine: areas of emerging state consensus on AI privacy; the consequences of polycentric compliance; and existing governance gaps. Our analysis reveals four persistent gaps in the U.S. regulatory architecture: a lack of explicit oversight regarding inferential analytics; limited remedies for algorithmic bias; the exclusion of emerging risks (e.g., deepfakes and synthetic media); and highly heterogeneous enforcement mechanisms that rely on varying degrees of regulatory and private oversight. These governance gaps contribute to a fragmented compliance environment that disproportionately impacts vulnerable populations and undermines accountability. This study ultimately highlights the urgent need for harmonized frameworks that transcend the limitations of current polycentric approaches to ensure equitable and accountable AI governance.
- Open Access
- 20 March 2026
- Published by University of Borås
- Vol. 31 (iConf) , 684-698
Competing visions of ethical AI: a case study of OpenAI
Introduction. AI Ethics is framed distinctly across actors and stakeholder groups. We report results from a case study of OpenAI analysing ethical AI discourse. Method. Research addressed: How has OpenAI’s public discourse leveraged ‘ethics’, ‘safety’, ‘alignment’ and adjacent related concepts over time, and what does discourse signal about framing in practice? A structured corpus, differentiating between communication for a general audience and communication with an academic audience, was assembled from public documentation. Analysis. Qualitative content analysis of ethical themes combined inductively derived and deductively applied codes. Quantitative analysis leveraged computational content analysis methods via NLP to model topics and quantify changes in rhetoric over time. Visualizations report aggregate results. For reproducible results, we have released our code at https://github.com/famous-blue-raincoat/AI_Ethics_Discourse. Results. Results indicate that safety and risk discourse dominate OpenAI’s public communication and documentation, without applying academic and advocacy ethics frameworks or vocabularies. Conclusions. Implications for governance are presented, along with discussion of ethics-washing practices in industry.
- 20 March 2026
- Published by University of Borås
- Vol. 31 (iConf) , 684-698
Open AccessCompeting visions of ethical AI: a case study of OpenAI
Introduction. AI Ethics is framed distinctly across actors and stakeholder groups. We report results from a case study of OpenAI analysing ethical AI discourse. Method. Research addressed: How has OpenAI’s public discourse leveraged ‘ethics’, ‘safety’, ‘alignment’ and adjacent related concepts over time, and what does discourse signal about framing in practice? A structured corpus, differentiating between communication for a general audience and communication with an academic audience, was assembled from public documentation. Analysis. Qualitative content analysis of ethical themes combined inductively derived and deductively applied codes. Quantitative analysis leveraged computational content analysis methods via NLP to model topics and quantify changes in rhetoric over time. Visualizations report aggregate results. For reproducible results, we have released our code at https://github.com/famous-blue-raincoat/AI_Ethics_Discourse. Results. Results indicate that safety and risk discourse dominate OpenAI’s public communication and documentation, without applying academic and advocacy ethics frameworks or vocabularies. Conclusions. Implications for governance are presented, along with discussion of ethics-washing practices in industry.
- Open Access
Competing Visions of Ethical AI: A Case Study of OpenAI
Introduction. AI Ethics is framed distinctly across actors and stakeholder groups. We report results from a case study of OpenAI analysing ethical AI discourse. Method. Research addressed: How has OpenAI's public discourse leveraged 'ethics', 'safety', 'alignment' and adjacent related concepts over time, and what does discourse signal about framing in practice? A structured corpus, differentiating between communication for a general audience and communication with an academic audience, was assembled from public documentation. Analysis. Qualitative content analysis of ethical themes combined inductively derived and deductively applied codes. Quantitative analysis leveraged computational content analysis methods via NLP to model topics and quantify changes in rhetoric over time. Visualizations report aggregate results. For reproducible results, we have released our code at https://github.com/famous-blue-raincoat/AI_Ethics_Discourse. Results. Results indicate that safety and risk discourse dominate OpenAI's public communication and documentation, without applying academic and advocacy ethics frameworks or vocabularies. Conclusions. Implications for governance are presented, along with discussion of ethics-washing practices in industry.
Open AccessCompeting Visions of Ethical AI: A Case Study of OpenAI
Introduction. AI Ethics is framed distinctly across actors and stakeholder groups. We report results from a case study of OpenAI analysing ethical AI discourse. Method. Research addressed: How has OpenAI's public discourse leveraged 'ethics', 'safety', 'alignment' and adjacent related concepts over time, and what does discourse signal about framing in practice? A structured corpus, differentiating between communication for a general audience and communication with an academic audience, was assembled from public documentation. Analysis. Qualitative content analysis of ethical themes combined inductively derived and deductively applied codes. Quantitative analysis leveraged computational content analysis methods via NLP to model topics and quantify changes in rhetoric over time. Visualizations report aggregate results. For reproducible results, we have released our code at https://github.com/famous-blue-raincoat/AI_Ethics_Discourse. Results. Results indicate that safety and risk discourse dominate OpenAI's public communication and documentation, without applying academic and advocacy ethics frameworks or vocabularies. Conclusions. Implications for governance are presented, along with discussion of ethics-washing practices in industry.
- Open Access
- 11 March 2025
- Published by University of Borås
- Vol. 30 (iConf) , 508-522
‘Sora is incredible and scary’: public perceptions and governance challenges of text-to-video generative AI models
Introduction. Text-to-video generative AI models such as Sora OpenAI have the potential to disrupt multiple industries. In this paper, we report a qualitative social media analysis aiming to uncover people’s perceived impact of and concerns about Sora’s integration. Method. We collected and analysed comments (N=292) under popular posts about (1) Sora generated videos, (2) a comparison between Sora videos and Midjourney images, and (3) artists’ complaints about copyright infringement by Generative AI. Analysis. We employed a thematic analysis to analyse the collected comments. Results. We found that people were most concerned about Sora’s impact on content creation-related industries. Governance challenges included the for-profit nature of OpenAI, the blurred boundaries between real and fake content, human autonomy, data privacy, copyright issues, and environmental impact. Potential regulatory solutions proposed by people included law-enforced labelling of AI content and AI literacy education for the public. Conclusions. Based on the findings, we discuss the importance of gauging people’s tech perceptions early and propose policy recommendations to regulate Sora before its public release.
- 11 March 2025
- Published by University of Borås
- Vol. 30 (iConf) , 508-522
Open Access‘Sora is incredible and scary’: public perceptions and governance challenges of text-to-video generative AI models
Introduction. Text-to-video generative AI models such as Sora OpenAI have the potential to disrupt multiple industries. In this paper, we report a qualitative social media analysis aiming to uncover people’s perceived impact of and concerns about Sora’s integration. Method. We collected and analysed comments (N=292) under popular posts about (1) Sora generated videos, (2) a comparison between Sora videos and Midjourney images, and (3) artists’ complaints about copyright infringement by Generative AI. Analysis. We employed a thematic analysis to analyse the collected comments. Results. We found that people were most concerned about Sora’s impact on content creation-related industries. Governance challenges included the for-profit nature of OpenAI, the blurred boundaries between real and fake content, human autonomy, data privacy, copyright issues, and environmental impact. Potential regulatory solutions proposed by people included law-enforced labelling of AI content and AI literacy education for the public. Conclusions. Based on the findings, we discuss the importance of gauging people’s tech perceptions early and propose policy recommendations to regulate Sora before its public release.
- 12 February 2025
- Published by Taylor & Francis
- p. 108-126
Neutrality or contextuality
Data have significant social impacts at the individual, community, and social levels. Data governance reflects the assemblage of strategies, norms, and rules developed formally and informally to structure and constrain the collection, processing, uses, and retention of data over time. When data governance is driven by financial interests or technological innovation premised on assumptions of neutrality, as opposed to locally relevant human-centered values, research indicates that governance is often ineffective or inequitable, echoing decades of social informatics work demonstrating that technologies are not neutral. This chapter tests this implication directly, exploring the impact of five longstanding social informatics principles—non-neutrality, reciprocal shaping, transparency, embedded values, and paradoxical uptake outcomes—on data governance outcomes in public sector data sets from California and Illinois. It concludes with a contextually responsive guide to sociotechnical data governance for solely public data lifecycles in practice, as well as a research agenda for future ethical sociotechnical data governance.
- 12 February 2025
- Published by Taylor & Francis
- p. 108-126
Neutrality or contextuality
Data have significant social impacts at the individual, community, and social levels. Data governance reflects the assemblage of strategies, norms, and rules developed formally and informally to structure and constrain the collection, processing, uses, and retention of data over time. When data governance is driven by financial interests or technological innovation premised on assumptions of neutrality, as opposed to locally relevant human-centered values, research indicates that governance is often ineffective or inequitable, echoing decades of social informatics work demonstrating that technologies are not neutral. This chapter tests this implication directly, exploring the impact of five longstanding social informatics principles—non-neutrality, reciprocal shaping, transparency, embedded values, and paradoxical uptake outcomes—on data governance outcomes in public sector data sets from California and Illinois. It concludes with a contextually responsive guide to sociotechnical data governance for solely public data lifecycles in practice, as well as a research agenda for future ethical sociotechnical data governance.
- Open Access
- 1 January 2025
- Published by Elsevier in SSRN Electronic Journal
Modeling Elements of Copyright with Large Language Models
Copyright, simply put, is a form of intellectual property that is primarily concerned with protecting original authored work that is tangibly expressed [1]. Despite a clear definition at face value, parties often disagree about what constitutes original work. The field of copyright law exists to resolve those disputes when they arise, largely through detailed examination of the disputed work; legal representatives from both parties attempt to argue that the disputed work either does or does not follow rules established for the use of copyrightprotected work (fair use). With respect to Large Language Models (LLMs), copyright violations, and how to address them, have attracted significant attention recently [2], with many proposed solutions attempting to either constrain models from producing copyright-protected images/text or perturb potential training data. However, what is missing from the current landscape is an examination of how LLMs can provide relevant copyright information to users, with the hope that such information can prevent the use of copyrightprotected materials more generally. We aim to address this gap via comprehensive evaluation of the ability of multiple LLMs to answer complex questions related to copyright law, through comparative analysis of multiple LLMs and fine-tuning using annotated hypotheticals.
- 1 January 2025
- Published by Elsevier in SSRN Electronic Journal
Open AccessModeling Elements of Copyright with Large Language Models
Copyright, simply put, is a form of intellectual property that is primarily concerned with protecting original authored work that is tangibly expressed [1]. Despite a clear definition at face value, parties often disagree about what constitutes original work. The field of copyright law exists to resolve those disputes when they arise, largely through detailed examination of the disputed work; legal representatives from both parties attempt to argue that the disputed work either does or does not follow rules established for the use of copyrightprotected work (fair use). With respect to Large Language Models (LLMs), copyright violations, and how to address them, have attracted significant attention recently [2], with many proposed solutions attempting to either constrain models from producing copyright-protected images/text or perturb potential training data. However, what is missing from the current landscape is an examination of how LLMs can provide relevant copyright information to users, with the hope that such information can prevent the use of copyrightprotected materials more generally. We aim to address this gap via comprehensive evaluation of the ability of multiple LLMs to answer complex questions related to copyright law, through comparative analysis of multiple LLMs and fine-tuning using annotated hypotheticals.
- Open Access
- 1 January 2025
- Published by HICSS Conference Office
Critically Comparing Global AI Regulations
Artificial Intelligence (AI) regulations, addressing associated rights, limits of use, issues of bias, and accountability mechanisms have been introduced or adopted by numerous national, regional, and municipal governments. This paper comparatively examines global governance via regulatory documents from 64 countries, as well as the European Union and African Union. Leveraging mixed methods, we evaluate: changes in proposed AI regulation over time, topical and document similarities in policy priorities, hedging and uncertainty that may impact enforceability, and whose interests are reflected in policy parameters. Results indicate: regulatory capture by industry in North and South America; decolonial dissent in Africa against emerging global consensus; and rhetorical and substantive similarities across nations. In addition to providing practical policy recommendations, this research applies multi-modal governance conceptualizations to AI and presents a research agenda.
- 1 January 2025
- Published by HICSS Conference Office
Open AccessCritically Comparing Global AI Regulations
Artificial Intelligence (AI) regulations, addressing associated rights, limits of use, issues of bias, and accountability mechanisms have been introduced or adopted by numerous national, regional, and municipal governments. This paper comparatively examines global governance via regulatory documents from 64 countries, as well as the European Union and African Union. Leveraging mixed methods, we evaluate: changes in proposed AI regulation over time, topical and document similarities in policy priorities, hedging and uncertainty that may impact enforceability, and whose interests are reflected in policy parameters. Results indicate: regulatory capture by industry in North and South America; decolonial dissent in Africa against emerging global consensus; and rhetorical and substantive similarities across nations. In addition to providing practical policy recommendations, this research applies multi-modal governance conceptualizations to AI and presents a research agenda.
- Open Access
- 16 October 2024
- Published by University of Illinois Main Library in Proceedings of the ALISE Annual Conference
University Governance for Responsible AI
With the democratization of access to artificial intelligence via large language models, such as ChatGPT, questions about AI ethics in education are at the forefront of educational policymaking. How do we teach students to engage with AI ethically in coursework, research, and careers? What uses of AI constitute violations of academic integrity? What norms need to be established about AI in higher education? This research evaluates official strategy and policy documents from a wide array of research and liberal arts schools in the U.S. via a combination of thematic and structured content analysis, to surface policy and ethics concerns and the institutional structures that shape restrictions, rights, and interventions, following assessment of interrupter reliability. Results highlight the concerns of universities and emerging approaches to intervene with respect to the appropriate use of AI in education. Policy recommendations are made based on the emerging strategies and norms identified via analysis.
- 16 October 2024
- Published by University of Illinois Main Library in Proceedings of the ALISE Annual Conference
Open AccessUniversity Governance for Responsible AI
With the democratization of access to artificial intelligence via large language models, such as ChatGPT, questions about AI ethics in education are at the forefront of educational policymaking. How do we teach students to engage with AI ethically in coursework, research, and careers? What uses of AI constitute violations of academic integrity? What norms need to be established about AI in higher education? This research evaluates official strategy and policy documents from a wide array of research and liberal arts schools in the U.S. via a combination of thematic and structured content analysis, to surface policy and ethics concerns and the institutional structures that shape restrictions, rights, and interventions, following assessment of interrupter reliability. Results highlight the concerns of universities and emerging approaches to intervene with respect to the appropriate use of AI in education. Policy recommendations are made based on the emerging strategies and norms identified via analysis.
- Open Access
- 3 October 2024
- Published by Wiley in Journal of the Association for Information Science and Technology
- Vol. 76 (1) , 289-325
Sociotechnical governance of misinformation: An Annual Review of Information Science and Technology (ARIST) paper
Misinformation is a complex and urgent sociotechnical problem that requires meaningful governance, in addition to technical efforts aimed at detection or classification and intervention or literacy efforts aimed at promoting awareness and identification. This review draws on interdisciplinary literature—spanning information science, computer science, management, law, political science, public policy, journalism, communications, psychology, and sociology—to deliver an adaptable, descriptive governance model synthesized from past scholarship on the governance of misinformation. Crossing disciplines and contexts of study and cases, we characterize: the complexity and impact of misinformation as a governance challenge, what has been managed and governed relative to misinformation, the institutional structure of different governance parameters, and empirically identified sources of success and failure in different governance models. Our approach to support this review is based on systematic, structured literature review methods to synthesize and compare insights drawn from conceptual, qualitative, and quantitative empirical works published in or translated into English from 1991 to the present. This review contributes a model for misinformation governance research, an agenda for future research, and recommendations for contextually‐responsive and holistic governance.
- 3 October 2024
- Published by Wiley in Journal of the Association for Information Science and Technology
- Vol. 76 (1) , 289-325
Open AccessSociotechnical governance of misinformation: An Annual Review of Information Science and Technology (ARIST) paper
Misinformation is a complex and urgent sociotechnical problem that requires meaningful governance, in addition to technical efforts aimed at detection or classification and intervention or literacy efforts aimed at promoting awareness and identification. This review draws on interdisciplinary literature—spanning information science, computer science, management, law, political science, public policy, journalism, communications, psychology, and sociology—to deliver an adaptable, descriptive governance model synthesized from past scholarship on the governance of misinformation. Crossing disciplines and contexts of study and cases, we characterize: the complexity and impact of misinformation as a governance challenge, what has been managed and governed relative to misinformation, the institutional structure of different governance parameters, and empirically identified sources of success and failure in different governance models. Our approach to support this review is based on systematic, structured literature review methods to synthesize and compare insights drawn from conceptual, qualitative, and quantitative empirical works published in or translated into English from 1991 to the present. This review contributes a model for misinformation governance research, an agenda for future research, and recommendations for contextually‐responsive and holistic governance.