The first comprehensive look at how local councils across Australia are deploying artificial intelligence, what's actually live, where governance is keeping up, and what we can learn from 15 international examples across 5 continents.
90
NSW councils running AI tools
Out of 128 — Audit Office, 12 March 2026
109
Distinct AI tools in use
Pilot or production across NSW LGAs
11%
Have an AI adoption strategy
40% have any formal AI policy
$2.7M
Awarded under Early Adopter grants
16 NSW councils — $200k–$500k each
The first comprehensive audit of AI use across NSW councils — and the data point that frames the whole landscape.
90 of 128
Councils running AI tools
109
Distinct AI tools in use
11%
Have an AI adoption strategy
10%
Maintain a centralised AI inventory
40%
Have a formal AI policy
51%
Considering / planning more AI tools
Recommendation 1 — DPHI mandatory framework by 30 June 2026
The Audit Office has formally recommended that the Department of Planning, Housing and Infrastructure work with relevant agencies to set a mandatory AI framework for NSW councils by 30 June 2026. Council leaders should be ready for it.
The new evaluation rail councils will be measured against.
NSW Office for AI launched a refreshed Assessment Framework, built with CSIRO Data61.
Replaces subjective self-assessments with a standards-aligned approach.
Aligned with the Commonwealth National AI Assurance Framework and the EU AI Act.
Designed to identify the right level of oversight in <30 minutes for low-risk systems.
$3M total pool, $2.7M awarded to 16 councils. Grants up to $200k individual / $500k joint.
Confirmed recipients include Bayside, Blacktown City, Burwood, Cessnock City and Canterbury-Bankstown, plus 11 more on the Department of Planning portal.
Suppliers on the AI Solutions Panel: Adaptovate (DAISY), Archistar, PropCode CDC and others.
The denominator for any national rollout.
128
NSW
79
VIC
77
QLD
137
WA
68
SA
29
TAS
17
NT
1
ACT
Total ~537 councils nationally (ALGA figure; up to 547 cited including Indigenous councils).
Synthesised from the AU and global research streams.
PATTERN #1
Chatbots work for high-volume transactional queries; fail for normative guidance.
Buenos Aires Boti succeeds. NYC MyCity fails. Mason City fails. Rule: deflect routine, escalate normative.
PATTERN #2
Algorithmic transparency is necessary but not sufficient.
Helsinki and Amsterdam both publish registers; Amsterdam's welfare AI still discriminates. Rule: register + audit + outcome monitoring + willingness to abandon.
PATTERN #3
Centralised, interoperable infrastructure beats council-by-council silos.
Singapore Ask Jamie at 90+ agencies. Estonia Bürokratt as standard. Rule: AU sector bodies (LGNSW, MAV, LGAQ) should own shared platforms.
PATTERN #4
Public legitimacy and democratic governance outweigh technical sophistication.
Toronto Sidewalk Labs collapsed. Barcelona Decidim succeeds. Rule: participation > optimisation.
A concrete playbook combining the NSW Audit Office baseline, the national assurance framework, and the lessons that hold across 15 international examples.
Start with infrastructure vision
Pothole / asset detection from existing fleet vehicles is low-risk, high-ROI, and proven across Moreton Bay, Noosa and the Asset AI® cohort. It also generates clean data — the foundation for harder use cases.
Adopt the NSW AI Assessment Framework as the baseline
Replaces subjective self-assessments with a CSIRO-Data61-built, standards-aligned approach. Aligns with the Commonwealth National AI Assurance Framework and the EU AI Act. <30 minute oversight assessment for low-risk systems.
Pool procurement at sector-body level
Singapore Ask Jamie and Estonia Bürokratt show centralised infrastructure beats council-by-council silos. LGNSW / MAV / LGAQ should own shared platforms; the NSW AI Solutions Panel is a viable starting point.
Publish an AI register before deploying anything citizen-facing
Helsinki's register is the global standard. Newcastle's 'Golden Rule' is the AU principle-level equivalent. Mandatory under the NSW Audit Office's Recommendation 1 by 30 June 2026.
Avoid normative chatbot decisions
NYC MyCity ($600k, shut down) and Mason City book bans show LLMs can't be trusted for compliance decisions. Use chatbots to deflect routine queries; escalate everything normative to humans.
Multilingual + accessible from day one
Auckland's Te Reo / Pasifika first-class approach + Camden's WCAG 2.1 + 100-language auto-translation are the access bar. Anything less is a regression on equity.