Reading guide

    Research map

    The archive is built as a connected record rather than a stream. This map groups the published record by the question each record addresses.

    Section 1

    How organisations classify and govern AI

    Records in this group ask how AI is named, governed, and made legible to the institution responsible for it.

    How Large Organisations Classify AI in Public Governance Documents

    A Documentary Study of Fifteen Listed Enterprises

    Published working paper

    Published June 2026

    Governance and classification

    A documentary study of fifteen large publicly listed organisations examining where enterprise AI sits within public governance architectures, showing that AI is universally governed as software, security, or product risk rather than as an informational input to professional judgement.

    The Unread Instruction

    Published working paper

    Published July 2026

    System prompts and governance oversight

    A reading of nine public frameworks that guide boards and audit leaders on AI oversight. None require anyone to know about the organisation-authored instruction that controls what the AI produces: who wrote it, when it was last changed, or how it affects answers. The phrase "system prompt" does not appear in any document reviewed.

    The Mandate Study

    Published working paper

    Published July 2026

    Organisational pressure and adoption

    A public-record study examining how major enterprises move AI beyond optional tool availability through leadership mandates, performance reviews, incentives, workflow integration, and board governance.

    Section 2

    How regulation interfaces with human agency

    These records examine what law and compliance frameworks ask human operators to do, and where responsibility settles when governed systems still shape the decision.

    The Awareness Trap

    Article 14, Automation Bias, and the Gap Between Compliance and Protection Under the EU AI Act

    Published examination

    Published March 2026

    Regulation and human oversight

    An examination of Article 14 and Article 26 of the EU AI Act against empirical psychological research on automation bias, demonstrating that statutory oversight mandates demand cognitive capabilities humans cannot sustain during routine operation.

    The Liability Transfer

    Published working paper

    Published July 2026

    Responsibility and controls in enterprise AI contracts

    A reading of the public terms for seven enterprise AI services. Across all seven vendors, the contracts make the customer responsible for AI output. The same contracts describe controls that affect that output: filters, system instructions, safety features, retrieval rules, data masking, and classifiers. The clauses that assign responsibility do not mention those controls.

    Section 3

    What public reporting shows and omits

    This thread reads public corporate language to see what organisations say about AI, what they avoid saying, and how that prose changes over time.

    Adoption Without Capability Reporting

    How Large Organisations Disclose AI Adoption Without Reporting What the AI Can Do

    Published working paper

    Published June 2026

    Public reporting and disclosure

    A systematic coding study of 714 AI-related statements across twenty-two major firms examining how corporate management publicly describes AI programmes, showing that disclosures focus heavily on deployment scale, efficiency, and product marketing while direct measured contributions to business reasoning remain almost entirely unreported.

    The First Annual Reports of the LLM Era

    Language Change in 150 SEC 10-K Filings, 2019–2024

    Published working paper

    Published March 2026

    Public reporting and disclosure

    A longitudinal study of 150 SEC 10-K filings across 50 major US public companies comparing FY2019, FY2022, and FY2024, demonstrating a 24.5 per cent increase in the annual rate of prose drift and a shift toward diffuse hedging and framing drift following the introduction of enterprise LLM tools.

    The Executive Access Study

    Published working paper

    Published July 2026

    Executive access and disclosure

    A review of the public record revealing the significant corporate disclosure gap regarding exactly how senior executives at large organisations access AI capabilities.

    The Language of AI Deployment

    A Longitudinal Documentary Study of Nine Large Organisations, 2022 to 2026

    Published working paper

    Published July 2026

    Executive discourse and deployment claims

    A longitudinal documentary study of 155 verified public statements from nine major organisations between 2022 and 2026, tracking how corporate leaders publicly describe internal AI adoption across five core communication registers.

    Section 4

    What deployment changes in practice

    These studies examine the governed layer itself: what it removes, what it preserves, and whether anyone can step around it at the moment of use.

    The Subtraction Study

    Measuring Capability Reduction in Enterprise AI Configurations

    Published working paper

    Published June 2026

    Deployment and capability reduction

    A controlled empirical test evaluating whether the governance wrapping that makes a model deployable inside an enterprise systematically removes specific operational capabilities while preserving the superficial appearance of intelligence.

    The Override Rank Cannot Reach

    Published working paper

    Published July 2026

    Deployment controls and rank-insensitive override logic

    A documentary reading of seven enterprise AI deployments showing that public administration and compliance documentation does not recognise executive rank as a query-time basis for varying the governed layer.

    The Sparring Partner Study: How enterprise AI handles strategic decisions from senior executives

    Testing method

    Methodology locked in, 6 July 2026

    Method and grading rules

    The committed prompt set and scoring criteria for a comparative study of how different AI setups respond to strategic decisions from senior leaders. Five scenarios run through three setups each, producing fifteen answers scored blind on whether they take a clear position, add fresh insight, and serve as material a senior leader could debate. The methodology was locked before any outputs were generated.