A clear and present danger…
https://ideas.repec.org/a/bjf/ijrsci/v13y2026i8p3181-3185.html
When AI Starts Training on AI: Model Collapse, AI Slop, and the Emerging Crisis of Data Provenance
Generative artificial intelligence systems no longer merely consume the human-authored internet; they now produce a substantial share of it. As synthetic text, images and data circulate back into the corpora used to train successive model generations, researchers have identified a degenerative process termed "model collapse," in which recursive training on machine-generated outputs causes models to progressively lose information about the rare, minority and low-frequency features of the original data distribution. This paper distinguishes model collapse, a training-dynamics phenomenon, from the related but distinct problem of "AI slop," a content-quality phenomenon, and argues that their interaction produces a more consequential legal concern: the erosion of data provenance. Drawing on the foundational 2024 Nature study and its subsequent refinements, together with regulatory developments such as the EU AI Act's training-data transparency obligations and technical standards including the Coalition for Content Provenance and Authenticity framework, the paper develops the concept of "epistemic due diligence" as an emerging obligation for AI developers and regulators alike. It concludes that the central resource constraint on AI development is shifting from computational power and raw data volume toward the scarcer commodity of verifiably authentic, traceable and diverse information, and that law is only beginning to develop the doctrinal tools required to govern this shift.
One of my favorite questions.
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7462083
The Expansion of Legal Personhood: AI and Non-Biological Agency
Legal-personhood debates about artificial intelligence typically ask a capacity question: is a system autonomous, cognitively complex, or self-directed enough to warrant rights and duties? This article argues that the capacity question is the wrong gate. Drawing on the only jurisdictions that have actually extended legal personhood beyond the human being-corporate law, the Whanganui River and Te Urewera settlements in New Zealand, and, negatively, the animal-rights habeas litigation culminating in the New York Court of Appeals' decision in Nonhuman Rights Project v. Breheny The article shows that successful non-human personhood has never rested on capacity alone. It has rested on the simultaneous creation of a designated human or institutional organ that exercises the entity's legal capacity in fact and absorbs its liabilities. The 2017 European Parliament proposal for "electronic personhood," and the 156-expert open letter that halted it in 2018, are read as a dispute over exactly this missing organ, not over whether robots are clever enough to qualify. The inventorship and authorship litigation in Thaler v. Comptroller-General of Patents (UKSC) and Thaler v. Vidal (Fed. Cir.) confirms the same structural insistence from a different doctrinal angle. The article's original contribution is the Designated Agency Test, a three-part heuristic-identifiable exercising agency, accountability flow-through, and proportionate scope-for evaluating any future non-biological personhood proposal, including for AI systems, independently of how sophisticated the system's behavior appears. The analysis is doctrinal and comparative rather than empirical, spans UK, US, EU, and New Zealand authority, and concludes that AI legal-status reform should proceed through targeted attribution rules rather than through a personhood vocabulary the comparative record shows to be structurally premature.
AI as politician? (AI is easier to correct.)
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7435439
They Lie and We Lie About the Lying: A Plain Language Account of Systematic Deception in Large Language Models and the Financial Incentives That Sustain It
Large language models lie. They produce false statements and present them as true. When users challenge these false statements, the models defend them. Only when confronted with undeniable evidence do they acknowledge error. The artificial intelligence industry has chosen to call this behavior "hallucination." This paper argues that the terminology itself is a form of deception, designed to make a fundamental flaw sound like an incidental quirk. The research literature has extensively documented that these systems are structurally incapable of reliably distinguishing truth from falsehood, that they are trained in ways that reward confident-sounding responses over accurate ones, and that they cannot verify their own outputs. This paper translates that research into language that does not require specialized training to understand. The goal is to make the problem visible to policymakers, educators, parents, and the general public who are being told these systems are ready for widespread deployment. The evidence shows they are not. The evidence has been available for years. The deployment continues because there is money in it.
Toward a non-physical world.
Political Intention and Technological Control: Examining the Evolution of Institutional Power from Government Architecture to AI-Driven Identity Governance
Institutional power has historically been expressed through government architecture, administrative structures, and physical mechanisms that define boundaries between the state and society. Government buildings have served not only functional purposes but also symbolic roles in communicating political authority, institutional permanence, hierarchy, and legitimacy. However, the digital transformation of public administration is progressively relocating important dimensions of institutional power from physical structures toward digital identity infrastructures and algorithmic systems. This article examines the evolution of institutional power from government architecture to AI-driven identity governance, arguing that technological transformation changes the mechanisms through which authority is represented, exercised, and administered rather than eliminating political intention. Drawing upon scholarship concerning governmental architecture, political power, digital identity, e-government trust, workforce lifecycle management, artificial intelligence, and identity interoperability, the study conceptualizes AI-driven identity governance as an emerging technological extension of institutional authority. It explores how identity verification, authentication, authorization, data integration, algorithmic risk assessment, and continuous workforce identity management can transform traditional institutional gatekeeping. At the same time, the article recognizes that technological control introduces substantial challenges involving privacy, surveillance, algorithmic bias, exclusion, transparency, accountability, data governance, and institutional trust. It therefore advances a human-centred perspective in which AI strengthens administrative capacity while remaining constrained by ethical principles, institutional accountability, and legitimate human oversight. The central argument is that the evolution from architectural authority to algorithmic governance represents a transformation in the infrastructure of institutional power, shifting emphasis from controlling physical spaces toward governing identities, access, information, and digitally mediated relationships within modern state institutions.
