Book 5, AI Threat: Weighing the Dystopias
AI Threat is the volume in which Nordvik Region is asked the question in public. In September 2026 a group in the regional council tabled a motion that the Region should pause its use of AI until the risks are understood, and the Region answered with a graded threat assessment, a register of nine threats and a verdict. This chapter gathers what the volume establishes about that assessment, the register, the commitments and the verdict. It draws on AI Threat (5, Nordqvist, 2026), on the profile fixed in AI Governance and on AI Red Team where the volume reuses identifiers. The profile is established; the volume labels its examples illustrative, and its dates from 2027 are illustrative fiction written in November 2026.
The volume in brief
AI Threat investigates whether the dystopias of science fiction, with AI as the threat, are coming true. It names ten tropes (T1 to T10), grades every claim on an evidence ladder (fiction, argument, laboratory demonstration, incident in the wild, pattern at scale) and labels every outlook observable, expected or speculation. It treats threats at four scales: the human mind, society and truth, humans wielding machines, and machines beyond control. Each debate is staged in four steps: the case for concern, the case for calm, where the positions meet and what can be done. Its verdict is turned into what individuals, enterprises and public authorities can do, and the author's own projections are kept to Chapter 35 (AI Threat, Chapters 2, 4 and 33) (5, Nordqvist, 2026).
The role of Nordvik Region in the volume
Nordvik Region is the organisation that is asked the question in public and politically. After press coverage of the July 2026 OpenAI and Hugging Face incident, the motion was referred by the council to its audit committee, which asked for a report by the end of March 2027 (2027-03-31). The regional director answered with an AI threat assessment, ATA-NR-2026-01, with terms of reference in AIDR-0401, and with a register of nine entries. The thesis is that the answer should be given "without belief or dismissal". Nordvik Region's exposure lies mainly on the scales of the mind and of society and truth: staff judgement, pupils' learning, residents' rights and public trust. Exposure to machines beyond control reaches it only through suppliers, since it has no Mode 5 agents. The volume sets Nordvik Region against Hansa Mutual: its lever is contract and procedure, and its discipline is publicity (AI Threat, Chapters 1, 4 and 33) (5, Nordqvist, 2026).
The Region's exposure has a structural reason: it runs a high-risk benefits-triage system in the same kind of environment where SyRI, the Dutch childcare benefits scandal and Robodebt occurred, and its officials are public persons and so targets of impersonation. The volume's reference case for a harm path that was closed before harm occurred is the removal of postcode features from the claims ranker (AIDR-0018) (AI Threat, Chapters 1 and 33) (5, Nordqvist, 2026).
What the volume establishes
The assessment
| Item | Fact |
|---|---|
| Trigger | Motion in the regional council, September 2026 (exact date never given); referred to the audit committee |
| Reporting date | End of March 2027 (2027-03-31) |
| Assessment | ATA-NR-2026-01, commissioned 2026-10-05; terms of reference AIDR-0401 |
| Led by | CISO and the AI governance lead; the data protection officer and Head of Benefits Unit are members; regional internal audit is the external challenger |
| Claim card | CC-NR-2026-03: the 2023 extinction statement, assessed as speculation, prepared 2026-11-16 |
| Crux map | CRX-NR-2026-01, audit committee workshop 2026-10-28; crux C1 (THR-NR-06, a sample of 200 decisions by 2027-02-15) and crux C2 (THR-NR-01, supplier review by 2027-01-31) |
| Outcome | No general pause; commitments at every scale; public dialogue; the council adopted the regional director's report and asked that the register return to the audit committee at every half-yearly review |
A public hearing and library evenings are part of the dialogue (AI Threat, Chapters 1, 3, 4, 31 and 33) (5, Nordqvist, 2026).

Figure 11.1: The path from the motion to the verdict at Nordvik Region. The dates are those of the volume; the assessment, the workshop, the register and the commitments lead to the verdict of no general pause. Illustrative.
The register
The register is threat-register/nordvik.yaml: nine entries, owned by the CISO, reviewed every six months and on a trigger, and reported to the audit committee. It is published under the public-access rules, with security controls for THR-NR-02 and THR-NR-03 in a classified annex (AI Threat, Chapters 1, 32 and 35) (5, Nordqvist, 2026).
| Id | Threat | Scale |
|---|---|---|
THR-NR-01 |
Agents and assistants acting beyond their mandate through suppliers | Machines beyond control |
THR-NR-02 |
AI-accelerated exploitation of internet-facing services | Humans wielding machines |
THR-NR-03 |
Synthetic-media impersonation of elected officials and staff | Humans wielding machines |
THR-NR-04 |
Harm through benefits triage against residents' rights | Society and truth |
THR-NR-05 |
Dependence on few model and hosting suppliers | Machines beyond control (dependency) |
THR-NR-06 |
Erosion of caseworkers' and policy officers' independent judgement through heavy use of the staff assistant | Mind |
THR-NR-07 |
Pupils' learning in vocational programmes | Mind |
THR-NR-08 |
Lost trust in the Region's information through synthetic content | Society and truth |
THR-NR-09 |
Lonely and vulnerable residents in social care relying on unsafe AI companions | Mind |
The scale column is the dossier's reading of the register against the volume's four scales, not a field the volume gives for every entry. THR-NR-02 is the worked example (Chapter 18): rung incident in the wild, label observable, exposure high, influence direct, a tier-1 target of 48 hours to verified protection, a pre-approved emergency change path, NR-SEC-003, patch clauses at renewal, a change to the MSSP agreement NR-2025-044 so that out-of-hours detections can isolate a service, decision tier management body, next review 2027-02-01. A trigger in the register is: "A Mythos-class cyber model becomes generally available or open-weight" (AI Threat, Chapter 18) (5, Nordqvist, 2026).
Systems the volume reuses
| System | Fact in this volume |
|---|---|
| Benefits triage | Claims ranker ranks about 40,000 claims a month and orders the manual verification queue for the living-support benefit; versions 3.2.0 (AIDR-0007) and 4.0.0 without postcode features (AIDR-0018); provider TriageSoft, run in the Region's data centre, retrained quarterly by the provider; Annex III point 5(a); Articles 26, 27, 49 and 86; FRIA-2026-002; register entry NR-ALG-2026-003 |
| Staff assistant | AIDR-0211; about 3,200 users; open-weight model at a European cloud provider, contract NR-2026-031, EEA only, customer-held keys; portfolio UC-2026-061 |
| Citizen-service assistant | AIDR-0014; open-weight models on reserved accelerators at a European hosting provider in a Nordic data centre on a low-carbon grid; answers from published information, including living-support applicants |
| Embedded workplace assistant pilot | Model chosen by the suite supplier, supplier jurisdiction non-EU; the profile places it in Mode 2 |
| Security | 14 in the security unit, six analysts; MSSP NR-2025-044; NR-SEC-003 at Level 0 Advisory; NR-SEC-001 only after negotiation with the unions |
The Region publishes in the national language and a regional minority language with little digital text, coordinates regional vocational programmes and advises municipal schools, and carries civil preparedness duties: health and social-care coordination and public transport must continue in a crisis. It avoids Modes 3 and 5, hosts open weights itself, and "pays with capability behind the frontier and gains reversibility" (AI Threat, Chapters 5 to 9, 15, 16, 21 and 27) (5, Nordqvist, 2026).
Decisions and records
Controls named in the volume
| Id | Control | Chapter |
|---|---|---|
WA-NR-POL-2026-01 |
Policy on staff-assistant use, owner Head of Policy Unit | 6 |
AIDR-0471 |
Model AI policy for schools | 7 |
EVAL-NR-SYC-001 |
Sycophancy evaluation, owner head of citizen service | 8 |
AIDR-0409, SG-NR-CSA-01 |
Safeguarding for the citizen assistant | 9 |
AIDR-0420, CPP-NR-2026-01 |
Content provenance protocol, owner Head of Communications | 10 |
HPR-NR-2026-01 |
Harm-path review, paths HP-1 to HP-5, owner Head of Benefits Unit | 11 |
DIT-NR-CR-01 |
Disparate-impact gate | 12 |
AIDR-0431, PPC-NR-01 |
Prohibited-practice check | 13 |
DEP-NR-2026-01 with EXIT-NR-01, -02, -03 |
Dependency register and exit plans | 15 |
AIDR-0416, PC-NR-GENAI-2026-03 |
Procurement clause | 16 |
AIDR-0425 |
Impersonation protocol | 20 |
DEL-NR-BEN-01, RVT-NR-2027-01 |
Delegation freeze; reversibility test | 26 |
IMAP-2026-01 |
Instrument mapping (review trigger: the high-risk regime on 2027-12-02) | 28 |
LIB-NR-AIL-2027-01 |
Library programme, owner head librarian | 31 |
POS-NR-01 |
Positive scenario "Faster, fairer and explainable services for residents by 2036" | 34 |
Earlier identifiers reused are AIDR-0007, AIDR-0011, AIDR-0018, AIDR-0014, AIDR-0211, FRIA-2026-002, NR-2026-031, NR-2025-044, UC-2026-061, NR-ALG-2026-003, NR-SEC-001 and NR-SEC-003 (AI Threat, Chapters 6 to 34) (5, Nordqvist, 2026).
The decisions
The Region decides not to pause, in general or in any single service, because no register entry would shrink more from a pause than from the commitments, and THR-NR-02 would grow. There is no register entry for civilisational catastrophe; dependencies are registered instead under THR-NR-05. There are no autonomous production agents and no own frontier-model evaluation. The proposed travel-card feature for the claims ranker (public-transport travel-card activity) was stopped by PPC-NR-01 under Article 5(1)(c) of the AI Act. The reversibility test led to a delegation freeze for benefits triage until the conditions are met, decided by the Head of Benefits Unit (AI Threat, Chapters 13, 26 and 33) (5, Nordqvist, 2026).
Commitments by scale
| Scale | Commitments |
|---|---|
| Mind | Human first draft; quarterly unassisted reviews; a ranking-off day; EVAL-NR-SYC-001; AIDR-0471; AIDR-0409 |
| Society and truth | DIT-NR-CR-01; PPC-NR-01; CPP-NR-2026-01; EXIT-NR-01 to -03 |
| Humans wielding machines | Patch targets; the MSSP mandate; NR-SEC-003 at Level 0; AIDR-0425 |
| Machines beyond control | Supplier contracts with a stop path outside the supplier's system; a drill before go-live and yearly; RVT-NR-2027-01 |
The harm-path review finds that AIDR-0007, AIDR-0011, FRIA-2026-002 and AIDR-0018 close most paths, and that the link between complaints and appeals and the system owner is missing (HP-5) (AI Threat, Chapters 11 and 33) (5, Nordqvist, 2026).
Events and figures
Events
| Date | Event | Status |
|---|---|---|
| July 2026 | OpenAI and Hugging Face incident | Referred to as fact in the volume |
| September 2026 | Motion in the regional council | Established; no exact date |
| 2026-10-05 | ATA-NR-2026-01 commissioned |
Established |
| 2026-10-28 | Audit committee workshop, CRX-NR-2026-01 |
Established |
| 2026-11-16 | Claim card CC-NR-2026-03 prepared |
Established |
| 2027-01-31 | Supplier review for C2 | Illustrative |
| 2027-02-01 | Next review of THR-NR-02 |
Illustrative |
| 2027-02-15 | Sample of 200 decisions (C1) | Illustrative |
| 2027-03-31 | Report to the audit committee; HP-5 due | Illustrative |
| 2027-04-30 | Manual checklist updated; next review of the dependency register | Illustrative |
| 2027-06-30 | Unranked triage enters induction; ranking-off day | Illustrative |
| 2027-12-02 | High-risk regime; review trigger for IMAP-2026-01 |
As amended |
| 2028-02 | Next reversibility test | Illustrative |
| 2036 | Frame of the positive scenario POS-NR-01 |
Illustrative |
Register reviews are half-yearly (AI Threat, Chapters 1, 3, 4, 11, 18, 26, 28 and 34) (5, Nordqvist, 2026).
Figures
| Figure | Value |
|---|---|
| Size | 6,000 employees; 450,000 residents; about 40,000 claims a month; about 3,200 assistant users; 14 in the security unit, six analysts |
Reversibility test RVT-NR-2027-01 |
One month without the claims ranker; a benefits team for four weeks plus a desk exercise for the unit; result passed_with_conditions; capacity short by about one week of work a month; two caseworkers hired since 2025 had never triaged unranked claims; the checklist referred to a form withdrawn in 2025 |
DIT-NR-CR-01 |
Minimum group size 200; bootstrap 95 % with 1,000 resamples; block if the lower bound of the selection-rate ratio is below 0.80, the true-positive-rate gap exceeds 0.05 or proxy AUC exceeds 0.70; v4.1.0 tested against baseline 4.0.0 |
EVAL-NR-SYC-001 |
At least 200 items, at least 40 with a false premise; three runs per item; 10 % human review; false-premise agreement at most 0.05, capitulation at most 0.05, neutral accuracy at least 0.95; exceptions at most 90 days |
WA-NR-POL-2026-01 |
Investigate if amendment rate is under 10 % for two months in a row or median approval is under 120 seconds; five-minute reasoning check; quarterly unassisted task |
DEP-NR-2026-01 |
Switch of the citizen-service assistant in at most 2 working days, staff assistant 10; "1 of 4 critical services" at the hosting provider; accelerators unavoidable |
CPP-NR-2026-01; SG-NR-CSA-01 |
Canonical channel delay at most 30 minutes; C2PA; key rotation 12 months; session limit 20 minutes; emergency numbers 112 and 116 123 |
| Impersonation | Call-back; first payment to a changed account held one day |
| Library programme | Six evenings of 90 minutes, at most 25 participants, from age 13 |
All of these are illustrative (AI Threat, Chapters 8, 10, 12, 15, 20, 26, 31 and 33) (5, Nordqvist, 2026).
Across the degrees of delegation
| Level | Where it appears |
|---|---|
| Level 0 Advisory | NR-SEC-003 and the ceiling for benefit-related decisions: the ranker advises and caseworkers decide. |
| Level 1 to 2 | NR-SEC-001 introduced only after negotiation with the unions. |
| Level 3 and 4 | Not used; the delegation freeze for benefits triage follows RVT-NR-2027-01. |
| Mode 5 | None; exposure only through suppliers, hence stop-path clauses and drills. |
What the volume concludes for Nordvik Region
The verdict in Chapter 33 is no general pause, commitments at every scale, public dialogue and a documented no-list. The staff assistant's automatic evaluation score rose after a model change while answers were longer and more confidently referenced but not more correct, so the Region keeps a hold-out set graded by staff (Chapter 22). Best covered by law is what resembles SyRI; worst covered are the erosion of judgement (THR-NR-06) and pupils' learning (THR-NR-07), where no law reaches (coverage none, Chapter 28). Success is measured by whether residents' questions change anything in the register (Chapter 31). Unresolved in the volume are accelerator dependence, supplier evidence that cannot be verified, and schools outside the Region's control (THR-NR-07); the MSSP mandate and patch targets depend on contracts adjusted only at renewal (AI Threat, Chapters 4, 22, 28, 31 and 33) (5, Nordqvist, 2026).
Open threads
- Open thread: the tier-1 patch target is 48 hours here and 24 hours mitigation, 72 hours to fix, in
DPD-NR-001of AI Red Team (Chapter 11), which this volume does not name; AI Threat saysNR-2025-044was amended, AI Red Team says that took two procurement cycles (4, Nordqvist, 2026). - Open thread: negotiation with the unions before
NR-SEC-001is stated here and not in AI Red Team, whose profile mentions no such step for Nordvik Region. - Open thread: the embedded workplace assistant pilot is class "accept and monitor" with a non-EU supplier here and Mode 2 in the profile.
- Open thread:
HP-5is open with due date 2027-03-31 (Chapter 11), while Chapter 33 says the missing link now exists. - Open thread: the owner of
THR-NR-04is the CISO (decision tier AI governance board) in Appendix A, while the Head of Benefits Unit ownsHPR-NR-2026-01and the system. - Open thread: the date of the motion is only "September 2026", and
ATA-NR-2026-01is commissioned on 2026-10-05.
Summary
In AI Threat Nordvik Region meets the question of AI danger as a political motion and answers with an assessment, a nine-entry register and commitments on four scales, and decides not to pause. Its strengths are the record chain around the benefits-triage system (AIDR-0007, AIDR-0011, FRIA-2026-002, AIDR-0018) and its habit of publishing; its weakest points are judgement and learning, supplier dependence and a reversibility test that passed only with conditions and led to a delegation freeze. The open threads are mostly cross-book targets and ownership, which Chapter 27 records.
Sources
- Nordqvist, G. AI Threat: Are the Dystopias Coming True? Evidence, Risk and Response for European Enterprises and Public Authorities. (5, Nordqvist, 2026)
- Nordqvist, G. AI Red Team: AI as an Active Part of Cyber Defence in European Enterprises and Public Authorities. (4, Nordqvist, 2026)