City AI 360
Responsible intelligence at city scale.
A governance workspace for every AI system a city operates — registry, controls, monitoring and a public record.
NATIONWIDE AI STRATEGY · CITY DEPLOYMENT
The city AI portfolio
Every system registered, classified and owned.
AI system
Accountable entity
Risk tier
Assurance
Traffic incident prediction
Traffic Operations
L3 · High
Controlled
Citizen service assistant
Citizen Experience
L1 · Limited
Trusted
Clinical demand forecast
Health Planning
L3 · High
Controlled
Energy load optimisation
Utility Operations
L3 · High
Trusted
Planning document review
Planning Authority
L2 · Significant
Attention
Benefits eligibility triage
Welfare Authority
L3 · High
Intervention
Public space anomaly detection
City Safety Office
L4 · Restricted
Intervention
Public value and inherent risk are assessed on separate scales. A high-scoring use case with missing evidence does not reach production.
The governance operating model
Clear decisions. Named owners. Separated duties.
Four connected layers
01
City outcomes
Mobility, health, environment and citizen experience set the purpose.
02
Activation engine
Discover, prepare data, build, validate, deploy, monitor, retire.
03
Responsible AI controls
Ownership, data, fairness, security, oversight and independent assurance.
04
City intelligence
A registry and observability layer connecting change, incident and impact.
Who decides what
City AI Council
Sets priorities and risk appetite; approves high-risk deployment.
Responsible AI & Ethics Board
Independent review; signs the conformity decision separately from delivery.
City AI Office
Owns standards, registry, procurement guidance and monitoring.
Entity AI leads
Business, technical, data and risk owners accountable per system.
Responsible AI controls
Eight control families. One production gate.
GOV-01
Accountable owner
AI Act Art. 17
DAT-02
Data provenance
AI Act Art. 10
FAI-03
Fairness evaluation
ISO 42001 A.5.2
SEC-04
Adversarial testing
AI Act Art. 15
HUM-05
Human override
AI Act Art. 14
MON-06
Continuous monitoring
AI Act Art. 72
VEN-07
Vendor audit rights
AI Act Art. 25
AUD-08
Independent approval
AI Act Art. 43
G0 IdeaG1 DataG2 AssessmentG3 SandboxG4 ProductionG5 MonitorG6 Retire
The gate opens on recorded evidence and an independent sign-off. An open incident closes it again automatically.
Monitoring and the audit trail
Assurance continues after launch, or it never happened.
L1
Threshold alert
Drift crosses a documented limit; the model owner is paged within the hour.
L2
Owner triage
Impact assessed; fallback or reduced-autonomy mode applied immediately.
L3
Board escalation
High and critical incidents suspend the production gate automatically.
L4
Council and public record
Material harm is reported and summarised on the citizen register.
Hash-chained decision record
Registration, evidence, sign-off, incident and publication each write one entry, fingerprinted with the entry before it. Altering any past record invalidates every record after it.
041 · EVIDENCE.RECORD · AI-005 SEC-049f2c…a71d
042 · MONITOR.ALERT · AI-007 drift 7.4c30b…4e88
043 · INCIDENT.ESCALATE · INC-04117ad…90f2
044 · ASSURANCE.SIGNOFF · AI-004e6d1…2b45
Public accountability
What the city publishes, and how residents contest it.
AI-007 · SOCIAL SERVICES
Benefits eligibility triage
Support applications are ordered by likely urgency so caseworkers reach the most pressing cases first.
Human decision. Entitlement is always decided by a named caseworker.
Challenge it. Request a review and a full human reassessment.
01
Ask what happened
Plain-language explanation and the data categories used.
02
Human reassessment
A named officer re-decides without the system output in front of them.
03
Independent review
Unresolved cases follow the existing ombudsman route.
04
Systemic feedback
Upheld appeals count as a fairness signal in the next model review.
Activation
Three stages, then one governed exchange.
Prepare · Months 1–3
Establish the foundations
Executive mandate, a dedicated City AI Office, and the same four owner roles mirrored in every entity.
Execute · Months 4–6
Prove value in a sandbox
Rapid build–iterate–rollout cycles, one impact-driven framework, partnering balanced with capability build.
Sustain · Months 7–12
Operate, assure and scale
Feedback loops that reach the model, transparent impact reporting, and scale or retire on evidence.
The City AI Exchange
One admission rule for the whole city.
Data banks, approved models, use cases, sandbox and shared services sit in one exchange. A listing requires a classified risk tier, recorded control evidence and — at L2 and above — Ethics Board sign-off in the audit trail.
Success indicators
What the city measures, and what good looks like.
Public value
≥15%
Service outcome uplift
Per deployed system against its pre-deployment baseline.
40k
Officer hours redirected
Annual hours moved from routine processing to complex casework.
Delivery
≤6 mo
Idea to production
Median cycle time at L1–L2, gates included.
60%
Reuse across entities
Proven systems adopted by a second entity through the exchange.
Assurance health
100%
Evidence complete at L3+
No high-risk system in production with an outstanding control.
≤10 d
Incident closure
Working days from threshold breach to remediation signed off.
Public trust
≥85
Portfolio trust index
Weighted mean across nine documented dimensions.
100%
Resident-facing published
Every system touching residents on the citizen register, with an appeal route.
Targets are illustrative and set with the City AI Council at baseline. Value and delivery indicators say whether the programme is worth running; assurance and trust indicators say whether it is safe to keep running.
Framework
The City AI 360 framework
A proposed operating system for AI activation, responsibility and public trust — the city-scale instance of the S.AI.F assurance model.
The city AI charter
- Human centricity
- Fairness
- Transparency
- Explainability
- Accountability
- Privacy
- Safety
- Security
- Human oversight
- Contestability
Trust index composition
Each dimension is evaluated on a documented 0–100 scale, then weighted.
Safety
20%
Fairness
15%
Cybersecurity
15%
Transparency
10%
Privacy
10%
Human oversight
10%
Reliability
10%
Regulatory compliance
5%
Citizen trust
5%
92–100 Trusted · 80–91 Controlled · 65–79 Attention · below 65 Intervention. A critical failure overrides the aggregate score.
Five levels of city AI maturity
1 · Exploratory
Isolated pilots with limited common standards.
2 · Structured
A strategy and initial standards guide delivery.
3 · Governed
Registry, risk tiers and accountable governance are operating.
4 · Intelligent
Integrated monitoring connects the city-wide portfolio.
5 · Adaptive
Continuous assurance and citizen feedback inform the ecosystem.
Strategy, governance, data, technology, security, talent, procurement, monitoring and citizen participation are assessed separately. A maturity level must be supported by evidence.
The national context a city sits inside.
City-level governance has to answer to a national strategy. The published frameworks, with the point where S.AI.F controls attach to each objective, are mapped in the strategy library.