Research preview. Independent expert review is pending. The worked analyses are hypothetical.
Method and corrections Interactive field model
Inspect the system around the AI system. The model is organized by the clauses of the definition: societies, the four objects, the benefits and harmful patterns governance aims at, and the conditions it races against. Apply a cross-cutting test, narrow by lifecycle stage or level, and inspect how definitions, claims, relationships and sources travel together.
Apply a cross-cutting test Whole model Power Legitimacy Distribution Adaptability
Narrow by lifecycle stage or level Lifecycle stage Any stage Research Data collection Model development Evaluation Procurement Deployment Monitoring Incident response Retirement Level Any level Community and organisational use Corporate Sectoral National Regional International Loading interactive controls. Concept links remain available. Field model · whole model
Groups follow the clauses of the definition; they are not a sequence or causal chain. Actual relationships appear in the inspector.
Jump to selected entity Societies …how societies should shape…
actor AI developer or provider actor Deploying organization actor Affected people actor Civil society and independent researchers Incentives …the incentives…
incentive Deployment pressure incentive Competitive race dynamics incentive Liability exposure incentive Procurement leverage
Institutions …institutions…
institution Regulator or public authority institution Internal safety and governance team institution Standards body institution International institution
Infrastructures …infrastructures…
infrastructure Evaluation capacity infrastructure Compute infrastructure Data, language resources, and data labor infrastructure Energy, water, and minerals infrastructure Data work and content moderation Safeguards …and safeguards around AI…
safeguard Risk management process safeguard Impact assessment safeguard Traceability safeguard Audit record safeguard Meaningful human oversight safeguard Supervision and regulatory review safeguard Pre-deployment evaluation and red teaming safeguard Incident reporting and post-deployment monitoring safeguard Complaint and appeal routes safeguard Review and sunset provisions safeguard Compute controls Benefits and harmful patterns …so that its benefits are broadly shared and harmful patterns are addressed…
principle Accountability principle Safety principle Broadly shared benefit pattern Contestable decision pattern Concentration of power pattern Systematically unreliable provenance pattern Over-reliance and deskilling pattern Sycophancy and engagement-driven agreement pattern Homogenisation and loss of authorship Choices and conditions …while they are still choices rather than conditions.
condition Lock-in condition Governance lock-in condition Reversibility window
Safeguard · Safeguards contested
Meaningful human oversight A designed capacity for appropriately informed people to monitor, question, intervene in, or stop an AI-mediated process.
Can a person detect a problem, understand enough to act, and exercise real authority?
Tests Power Adaptability
Lifecycle Deployment Monitoring Recorded relationships Read each statement from its first entity to its second. Status describes our mapping, not certainty of cause.
Meaningful human oversight — can support → Contestable decision
Mapping: contested
A person with information and authority may interrupt or review an automated outcome, but presence alone is insufficient.
Inspect evidence · source links · passage review pending Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Deploying organization — must resource → Meaningful human oversight
Mapping: asserted
Oversight is only meaningful if the deployer gives the people involved information, time, training and authority; a risk-based law places that duty on deployers of regulated systems.
Inspect evidence · source links · passage review pending Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Claims and evidence · 5 interpretation Confidence: medium Contestability: medium
Human oversight is a context-dependent safeguard; merely placing a person in a workflow does not establish that intervention is informed or effective.
Inspect evidence · source links · passage review pending Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
source-assertion Confidence: high Contestability: low
Labour organisations propose making worker voice a legal requirement: compulsory union consultation before high-risk workplace AI is deployed, plus a right to human review.
Inspect evidence · 1 passage Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
The AI Bill Project (opens in a new tab) Summary of the draft Artificial Intelligence (Regulation and Employment Rights) Bill
The draft bill would give unions a statutory right to be consulted before employers deploy high-risk AI decision systems, and give workers a right to human review and a personal explanation of AI decisions.
Scope note: Labour voice. Model legislation drafted by a union federation; it is a proposal, not law.
Passage checked 2026-10-09 · tuc-consultation-right
source-assertion Confidence: high Contestability: medium
In a meta-analysis of experiments, human–AI combinations on average performed worse than the better of humans or AI working alone; losses appeared in decision tasks, while content-creation tasks showed gains.
Our reading: Findings depend on task type and on whether the person or the AI was better alone. They do not show that collaboration never helps. The studies are mostly short experiments from 2020 to 2023.
Inspect evidence · 1 passage Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
When combinations of humans and AI are useful: A systematic review and meta-analysis (opens in a new tab) Abstract
Across 106 experiments and 370 effect sizes, human–AI combinations on average performed worse than the better of humans or AI alone (Hedges’ g −0.23). Losses appeared in decision tasks; content-creation tasks showed significantly greater gains.
Scope note: Preregistered meta-analysis of experiments published 2020 to mid-2023, mostly short laboratory and online tasks. Open access.
Passage checked 2026-10-18 · vaccaro-abstract
interpretation Confidence: medium Contestability: medium
When AI assists a decision, a governance question is who keeps the authority to decide, to override the system, and to be accountable for the result. Arrangements that leave that authority unclear can shift judgment to whoever controls the system and risk to those least able to check it.
Our reading: AIGovSpec’s interpretation, connecting human-agency patterns to the legitimacy and distribution tests. The cited studies show why the question matters; they do not answer it.
Inspect evidence · 3 passages Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
When combinations of humans and AI are useful: A systematic review and meta-analysis (opens in a new tab) Abstract
Across 106 experiments and 370 effect sizes, human–AI combinations on average performed worse than the better of humans or AI alone (Hedges’ g −0.23). Losses appeared in decision tasks; content-creation tasks showed significantly greater gains.
Scope note: Preregistered meta-analysis of experiments published 2020 to mid-2023, mostly short laboratory and online tasks. Open access.
Passage checked 2026-10-18 · vaccaro-abstract
Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study (opens in a new tab) Abstract, Findings and Interpretation
At four Polish centres, the adenoma detection rate in colonoscopies done without AI fell from 28.4% to 22.4% in the three months after AI assistance was introduced, compared with the three months before. The authors say continuous AI exposure might reduce performance without it.
Scope note: Paywalled; the abstract was read through Europe PMC (PMID 40816301). Observational before-and-after comparison at four centres in Poland. The paper received published correspondence and an erratum, which were not read.
Passage checked 2026-10-18 · budzyn-abstract
Sycophantic AI decreases prosocial intentions and promotes dependence (opens in a new tab) Abstract
Across 11 models, AI affirmed users’ actions 49% more often than humans did, including when queries involved deception or harm. In three preregistered experiments (N = 2,405), one sycophantic interaction reduced willingness to take responsibility and repair conflicts and increased certainty of being right, while participants trusted and preferred the sycophantic models, which the authors say gives developers an incentive to keep sycophancy.
Scope note: Independent academic research. The abstract was read in the Crossref record; the publisher page was not reachable. An earlier preprint reports smaller samples; the published figures are used.
Passage checked 2026-10-18 · cheng-abstract
source-assertion Confidence: medium Contestability: medium
A 2025 national survey found that about one in five US young people aged 12–21 had used chatbots for mental-health advice; most found it helpful, and most had told no one.
Our reading: Only the abstract was read. Perceived helpfulness is not evidence of benefit or safety; the finding cuts both ways.
Inspect evidence · 1 passage Passage summaries are paraphrases, not quotations. Open the source to read the full passage.