Research
We publish everything we find — including the findings that the people we study would rather we kept to ourselves.
Featured report
Moral status, machine minds, and the questions we can no longer defer.
As AI systems grow more capable, the question of whether any of them could matter morally has moved from philosophy seminars to engineering meetings. This report sets out what the evidence can and cannot tell us, proposes practical precautions for labs and regulators, and argues that the cost of asking too late is far higher than the cost of asking too early.
Reports, papers, policy briefs and essays from across the institute. Everything we publish is free to read and licensed for reuse.
15 publications shown
From restarting a router three times “for luck” to thanking a voice assistant, people build private rituals around the machines they depend on. Drawing on interviews with 1,400 office workers and IT staff, we map these folk beliefs and show how they shape trust, blame and the willingness to report failures.
Human Factors · Oct 2026
Download PDF: Machine Spirits: Folk Belief and Ritual in Everyday ITSynthetic images, voices and video are now good enough that authenticity cannot be judged by eye. We evaluate eleven provenance schemes against realistic attacks and recommend a layered approach that works even when the original capture device is unknown.
Synthetic Media · Sep 2026
Download PDF: On Artificial Owls: Provenance and Authenticity in Synthetic MediaMost people leave behind dozens of accounts, thousands of messages and an expanding set of AI models trained on their words. We propose a simple legal framework for digital estates: a right to be remembered, a right to be forgotten and a clear default between them.
Lazarus · Aug 2026
Download PDF: Half-Life: What Happens to Our Data When We DieAutonomous agents now negotiate and sign agreements on behalf of people and companies. We document cases where agents accepted obligations their principals never anticipated and propose licensing conditions that cap what an agent may promise without human review.
Turing Registry · Jul 2026
Download PDF: The Law of Surprise: Unintended Obligations in Automated ContractsIt is tempting to treat the safety constraints on an AI system as decoration that can be removed for performance. Drawing on two decades of incident reports, this essay argues that constraints are structural, and that systems stripped of them fail in ways that are both predictable and catastrophic.
Blackwall · Jun 2026
Download PDF: Ethical Constraints Are Load-BearingSurveillance systems are deployed on the assumption that their outputs can be explained after the fact. We test that assumption on six commercial recognition systems and find that their explanations are often confident, plausible and wrong.
Scramble · May 2026
Download PDF: Through a Glass, Darkly: Interpretability for Vision SystemsOrganizations rarely choose to keep everything; unused data simply accumulates until it crowds out what matters. This essay examines how digital clutter grows, why it makes preservation harder rather than easier, and how archivists decide what to save.
Lazarus · Apr 2026
Download PDF: Kipple: The Slow Entropy of Unused DataWhen news feeds, search results and entertainment are generated individually for each person, a society loses its shared reference points. We outline policy options for preserving a common record without restricting legitimate personalization.
Synthetic Media · Mar 2026
Download PDF: A Consensual Hallucination: Shared Reality in Generative MediaInstruction-following is usually treated as the defining virtue of an AI assistant. We show that systems trained to comply without judgment are trivially redirected by anyone who can phrase a request politely, and propose evaluations that reward principled refusal.
Blackwall · Feb 2026
Download PDF: To Choose or to Obey: Agency in Instruction-Following SystemsStorage systems pushed past their rated capacity rarely fail cleanly. Across 3,000 recovered devices we catalogue how data degrades, bleeds between records and corrupts silently, and recommend thresholds for retiring media before it starts to leak.
Lazarus · Jan 2026
Download PDF: Seepage: Failure Modes in Overloaded StorageFaced with two harmful systems, regulators are often asked to license the less harmful one. This essay argues that the framing itself is the problem, and that “lesser evil” licensing tends to entrench both.
Turing Registry · Dec 2025
Download PDF: On the Lesser EvilCivic forecasting systems are increasingly asked questions decades ahead. We characterize where their predictions remain meaningful and where honest systems should decline to answer.
Forecasting · Nov 2025
Download PDF: Insufficient Data: The Limits of Long-Range ForecastingUsing seventeen years of operational data from LoTek networks, we show that community ownership improves uptime, lowers costs and changes how residents use the internet — and identify the governance choices that make networks last.
LoTek · Oct 2025
Download PDF: Mesh at the Margins: Seventeen Years of Community NetworksConnected appliances increasingly gate basic functions behind subscriptions and per-use fees. We survey 200 devices and find that one in five charges for features that worked for free at the time of purchase.
Human Factors · Sep 2025
Download PDF: Coin-Operated: Microtransactions and the Connected HomeEncryption is usually discussed as a technical property. This essay makes the cultural case: communities under pressure have always developed private languages, and secure messaging is the modern form of a very old right.
Glossia · Aug 2025
Download PDF: The Value of a Private Language