How AI should be regulated? Insights from the AI Regulation Early Career Researchers Conference 2026

The first day began with a welcome speech from Martin Kretschmer, CREATe Director, who introduced his ongoing research with Amy Thomas (University of Glasgow) on the AI licensing economy. This project analyses whether we are heading towards a licensing economy by collecting and analysing information on all “known commercial agreements” between content providers (licensors) and AI developers (licensees). The research maps the licensing terms across sectors and modalities, identifying some important patterns, such as the sub-sectors in which these agreements are more prevalent.

In her keynote “Neutrality, Fairness, Innovation, or Safety: what are we regulating AI for?”, Silvia de Conca (Vrije Universiteit) addressed a fundamental regulatory question: “what could (should?) we regulate AI for?” Her contribution set the tone for the conference by foregrounding the regulatory purposes and normative choices underlying some of the laws regulating AI at both Council of Europe (CoE) and European Union (EU) levels, including the ECHR, Convention 108+, the EU Charter, the AI Act, the DSA, the GDPR and the DMA. De Conca questioned the principle of technological neutrality and reflected on the role of the regulator. In her view, the regulator should be the one with normative power, with expertise grounded mainly in legislative and regulatory experience rather than technological experience.  She also expressed concerns about what she described as the ghost of “human rights” neutrality, arising from a fear of losing innovation.

The second keynote “Economic rationales behind AI litigation and policy”, was delivered by Xiaoren Wang (Dundee University), who analysed AI policy and litigation in the US, EU and China from an economic perspective, drawing on the Coase Theorem and related economic theories. She discussed, for example, the application of the four factors of the US fair use doctrine to the use copyright-protected works for AI training, considering the value created by both defendants and plaintiffs and arguing that the fair use decision essentially involves a value/harm evaluation. Wang applied this analysis to recent US fair use cases, particularly Bartz v. Anthropic (2025) and Kadrey v. Meta (2025). She also extended this economic analysis to EU policy, focusing on Article 4 of the CDSM Directive, as well as cases such as Gema v. OpenAI (2025) and Like Company v. Google (C-250/25) cases. Turning to China, she examined Article 7 of the Generative AI Regulation and the Ultraman case (2024), identifying some potential (although still inconclusive) consequences to AI litigation and copyright policy.

Chaired by Qingqin Zhang (University of Glasgow), the first panel focused on the intersection of AI, labour law and data protection law, with presentations exploring the growing use of algorithmic management and workplace monitoring technologies.

Ines Neves (University of Porto and CIJ – Centre for Interdisciplinary Research on Justice) presented her paper “Beyond Risk Classification: Worker-Centred Governance of AI in Industrial Workplaces”, which proposes to move beyond a purely risk-based and product-safety-oriented reading of the AI Act, by focusing instead on the governance of legitimate workplace uses of AI. Using industrial management and ergonomics as case studies, Neves argued that AI systems deployed to optimise logistics, occupational safety, ergonomics and worker well-being pursue objectives that warrant regulatory treatment beyond prohibition or rigid risk classification.

Tomasz Mirosławski (University of Miskolc) presented his research “Employee Monitoring as a Key Element of Technological Subordination within the Employment Relationship”, which critically analyses the notion of traditional employee subordination, particularly in the context of technological subordination related to algorithmic management, where constant software-based monitoring and control replace the instructions of a supervisor. By transforming the working environment into analysable digital data, the use of these technologies raises important questions about their compatibility with the Polish Labour Code and, more broadly, with the EU personal data protection regime. Mirosławski also addressed these issues through the lens of employee dignity.

Neil Saddington (University of Glasgow) presented his research “Refracted Transparency: Transparency as a Mechanism of Control”, which seeks to conceptualise the role that workplace transparency and the right to information can and should play in constraining the decision to engage in algorithmic management. The presentation focused on how the performative nature of workplace transparency and the right to information can be abused and mediated to benefit employers. Drawing on “performative transparency” theory as applied to the workplace and an analysis of Article 15 of the GDPR, Saddington highlighted the risk of transparency being used as a mechanism of control in the workplace.

José Miguel Diéguez Rodríguez (University of Murcia, Chair Integra Foundation on Identity and Digital Rights) presented his paper “Unsettled Categories: Neurodata-Driven Algorithmic Management at the Boundaries of Data Protection Law”, which assesses how AI-enabled workplace neurotechnologies used for occupational safety and health challenge the stability and internal coherence of core GDPR classifications (e.g., health data under article 9) when applied to data-intensive safety management. By testing whether existing legal categories can accommodate neurodata-driven operational control without conceptual slippage, the research identified a persistent uncertainty regarding the legal classification of neurodata in specific contexts and an absence of explicit national legal frameworks authorising its use. Rodríguez suggested that the proposal under the Digital Omnibus Directive may represent a significant step forward in addressing these gaps.

Chaired by Weiwei Yi (University of Glasgow), the second panel addressed AI in the context of consumer protection law with presentations covering a diverse range of topics, from manipulative and deceptive designs to transparency and explainability duties to product safety.

Beny Saputra (Central European University – CEU) presented his research “AI Robo-Advisers and Dark Patterns: Preventing Mis-Profiling and Unfair Personalisation”, which examines how AI robo-advisers on digital investment platforms use automated and data-driven systems, including AI, to recommend or manage investments, and analyses the risk of consumer manipulation posed by these systems through digital choice architecture and AI-enabled personalisation (“dark patterns”). Saputra discussed the legal implications of this and how the current legal framework could address these robo-advice dark patterns, highlighting the current legal fragmentation and proposing regulatory solutions.

Amanda Horzyk (Centre for Doctoral Training in Designing Responsible NLP, University of Edinburgh) presented her research “Engaging with Technical Communities on matters of Transparency and Explainability Law and Regulation: Translation, observations and aspirations drawn from practice”, in which she explores how the law affects and inspires practice, and interacts with AI systems design. Her presentation covered the current legal frameworks governing AI in the UK, the EU and internationally, including the AI Act, the GDPR, UNESCO Recommendations on the Ethics of AI, OECD AI principles and UNGA Council Resolutions. Horzyk also highlighted the challenges faced by the R&D XAI community when translating legal principles, such as Articles 15 and 22 of the UK Data Protection Act, into practice. The presentation focused on the particular difficulty of incorporating transparency and explainability principles into system design and proposed some open questions and ideas for moving forward.

Rebecca Owens (Durham University) presented her paper (co-authored with colleagues) “From Automation to Accountability: Designing Fair and Responsible Industrial Digital Twins”, in which they investigated the fairness and accountability challenges arising from autonomous industrial digital twins by drawing on the empirical data from a case study with an aluminium recycling partner. The presentation showed how autonomy and oversight, liability, transparency, automation bias, and workforce fairness intersect in digital twin governance and the existing structural tension given the epistemic opacity inherent in these systems. Owens highlighted that this conflicts with EU AI Act requirements for explainability, traceability, and human oversight. She also presented their proposal for a socio-technical framework to operationalise fairness and accountability principles within autonomous optimisation architectures.

Chaired by Aline Iramina (University of Glasgow), the third panel explored the relationship between AI and platform regulation, addressing issues such as AI health platforms, digital creative ecosystems, AI systems embedded in social media platforms and the development of AI auditing meta frameworks.

Federico Carmelo la Vattiata (University of Catania) presented his paper “Strong Safeguards and Agile Governance: Comparing EU and UK Approaches to Regulating AI Platforms”, which focuses on the core tension between prescriptive safeguards, which risk stifling innovation, and light-touch regimes, which risk enabling harm on a large scale. During his presentation, he compared the regulatory frameworks of the UK and the EU, arguing that their differences in ambition, structure, and enforcement could have significant implications for cross-border AI governance. Using AI-enabled healthcare platforms as a case study, he raised some key regulatory questions and challenges, such as those involving structural governance in data flows.

MingZhu Zhang (University of Galway) presented her research “Invisible by Design? Algorithmic Bias, Cultural Rights and Platform Regulation in Digital Creative Ecosystems”, which analyses the threat that digital platforms pose to cultural diversity, focusing on content moderation and other algorithmic systems that influence the visibility and engagement across the platforms and determine how creators reach their audience. Her presentation highlighted the opacity and centralisation of these systems, which greatly impact marginalised creators, as platforms prioritise popular content from a small number of creators. Zhang also covered some regulatory efforts, such as the CDSM Directive and the DSA, but argued that the practical impact of these remain uncertain. It concluded by advocating for an inclusive and rights-based platform governance.

Sejal Chandak (University of West of England) presented her paper “Can we have platform regulation without AI regulation?” , which examines the rapid integration of AI-driven text generation tools into social media platforms, a development that poses many risks and challenges, including for example the  proliferation of misinformation and manipulation, and an increase in violence against women and children. The presentation covered the Online Safety Act and considered whether chatbots and generative AI could represent the next stage in the evolution of algorithmic systems. It provided initial recommendations for the government, service providers, and civil society.

Derya S Esen (Goethe University Frankfurt) presented her paper “From Principles to Controls: Justifying AI Auditing Meta-Frameworks in Fragmented Regulatory Regimes”, which explores how consolidated AI auditing frameworks can be normatively justified, rather than appearing as ad hoc or purely technocratic compliance tools. The presentation outlined the context of the research, noting that prior to 2023, dozens of ethics guidelines were in place, offering high value but little auditability, enforcement or interoperability. However, since 2023, there has been a rapid proliferation of new AI laws, standards, and frameworks. Esen explained the research methodology, which was based on systematic reviews and meta-analyses, and presented the results of the scoping review. She highlighted the main goal of all initiatives and the common features, as well as the points of inconsistency and disagreement, and the blind spots, across frameworks. She also presented the normative design principles identified in the research that should guide the construction of meta-frameworks to ensure they maintain epistemic legitimacy.

Chaired by Gabriele Cifrodelli (Maastricht University), the fourth panel focused on the implications of AI for intellectual property law, exploring questions of originality and authorship in relation to generative AI outputs in the UK, the future of moral rights in AI‑driven film production and platform‑dominated creative markets, the power dynamics shaping EU text and data mining and copyright negotiations, and the role of AI within an increasingly complex patent system.

Cody Rei-Anderson (Edinburgh Napier University) presented his paper “Reconsidering copyright subsistence in generative AI outputs in the United Kingdom: Towards an originality analysis”, which addresses the question of the copyrightability of AI works amid the growing number of AI-generated and -assisted images, text, and video, arguing that how this question is resolved will have a profound impact on the UK’s creative economy. During his presentation, Rei-Anderson suggested that, rather than focusing on authorship, it would be more beneficial to analyse whether a work should be protected by copyright based on the closely related question of originality.  This legal concept should also be used to determine the conditions under which works created with or by AI could qualify for protection. He proposed a method for assessing when copyright might subsist in prompted AI-generated works, and concluded his presentation by outlining directions for future research.

Panagiotis Lampropoulos (University of Glasgow) presented his research “AI films and streaming giants: can moral rights save cinema?”, which  examines the production models pursued by big tech streaming companies and analyses them alongside empirical evidence demonstrating current knowledge of the uses and challenges of AI in creative industries, in order to establish the associated challenges. Lampropoulos also presented some of the findings using film and copyright theory to justify legal intervention through stronger moral rights.

Elodie Migliore (CEIPI, Université de Strasbourg) presented her research “Power Dynamics and Stakeholder Influence in the Negotiation of EU Text and Data Mining”, which  investigates the most influential actors in shaping the EU text and data mining exceptions, particularly Article 4 of the CDSM Directive, and explores how power relations between scientific research actors, commercial stakeholders, rightsholders, Member States and EU institutions affected the final regulatory outcome, and the extent to which AI was identified as a concern during the negotiations and how it was considered. Migliore presented the methodology, methods and data used in her research, as well as some of the main findings. For example, she highlighted the impact of the ministries responsible for IP in negotiations at the Council level, and the overall opacity of the procedure. She also emphasised that the AI discussed during negotiations was a different type of AI, underscoring the evolving nature of the concept.

Louise Jane Andrew (University of Glasgow) presented her paper “AI – A necessary evil in a complex patent system?”, which explores the extent to which AI is necessary to navigate complex global patent systems, due to multi-layered technological advancements, and whether it provides an opportunity for better access to patent information, reinvigorating the basis of the patent social contract. During her presentation, Andrew discussed how big data has the potential to transform the public sector and how AI could help to address some of the uncertainty in both patent boundaries and the (broken) patent social contract. She also addressed the risks of AI use and the current regulation, while highlighting the necessity and perhaps even the inevitability of AI use, as ‘a necessary evil’ in patent systems, considering the volume and complexity of data.

A Q&A session followed each panel, facilitating crosscutting discussion and peer feedback.

The second day of the conference was dedicated to an empirical methodology workshop led by Kristofer Erickson (University of Glasgow) and Amy Thomas. The workshop created space for participants to reflect on empirical and interdisciplinary methods in AI regulation research. The workshop introduced empirical methods for researching AI regulation and provided participants with practical guidance. They had the opportunity to work individually and in groups to design small research projects, receiving feedback from mentors and guests (Silvia de Conca (Vrije Universiteit) and Christos Christodoulopoulos (ICO)) in a safe and collaborative environment.

Participants proposed research projects focusing on AI and copyright law in movies and games, as well as AI transparency and explainability in the workplace context within the EU. Some participants have already expressed their interest and initiated the work on developing the paper together and creating a broader academic and interdisciplinary network on “Transparency and Explainability”.

By centring ECRs voices and providing a supportive space for work in progress, the AI Regulation ECR Conference 2026 demonstrated the value of creating dedicated scholarly forums at a moment when AI regulation is rapidly evolving at both national and international levels. Across panels, participants discussed shared challenges such as regulatory fragmentation, the tension between innovation and rights protection, and the difficulty of translating legal principles into operational AI governance.

Building on these discussions, CREATe will be launching a dedicated blog series, AI Regulation: ECRs’ Perspectives, featuring contributions from some of the early career researchers who presented at the conference. Each post offers an accessible insight into the research shared during the panels, allowing authors to develop their ideas further, reflect on feedback received, and situate their work within the rapidly evolving landscape of AI regulation.

The series will open on Thursday, 07 May with the blog post “The Performance of Transparency under Article 15(1)(h) UK GDPR” by Neil Saddington, in which he discusses how algorithms are infiltrating all stages of the employee lifecycle and analyses the role of transparency as an effective regulatory response to this. Further posts will follow in the coming weeks as part of this blog series. Readers are invited to follow the CREATe blog for upcoming contributions.

To receive updates on CREATe blog and research, subscribe to our newsletter or follow us on LinkedIn or BlueSky.