Essential services, population vulnerability and institutional capacity would score a council the same way for a natural disaster as for an AI-caused one. They measure who absorbs any shock badly, an approach inspired by the Australian Disaster Resilience Index (ADRI). We build on the legitimacy of the ADRI with an AI framing on top.
Of the six scored components, synthetic warning is the only one whose scored factor genuinely depends on AI as the cause, a fabricated AI-generated warning message during a live emergency. Deployment evidence and systemic labour are about AI specifically, but they measure adoption and exposure, not a mechanism by which AI itself causes harm.
In 2025 the Federal Court declined to find that the Commonwealth owed Torres Strait Islanders a duty of care over climate change harm. This project's legal grounding (see Legal background) rests on a narrower, different claim: that an AI-caused failure of a specific essential service is a concrete, attributable "event" under s16(1)(d) of the Disaster Management Act, not that government owes a general duty to prevent diffuse, hard-to-attribute harm. Climate-change causation is diffuse across decades and emitters; an AI-caused water or power outage in a named council area is not. Pabai is still worth naming, since Aurukun and the Torres Strait councils sit at the highest-risk end of this index: a court has already declined, in a case involving these same communities, to extend an affirmative duty over harm resembling this project's broader premise, even though the narrower statutory hook this project actually relies on is not the claim that case rejected.
Rogue AI, an AI system developing its own goal and resisting correction, requires a level of autonomous capability. Although this is a plausible catastrophic risk of AI, we were unable to find high quality data to represent this category and thus did not include it in the taxonomy.
Only 8 of 78 councils have confirmed AI use. This information was found by searching for news coverage. The other 70 are not confirmed to be free of AI use. We just were unable to find any publicly available information to say otherwise. The index treats "no evidence found" as missing data, not as evidence of absence, and excludes those councils from that component's score rather than scoring them as safe.
Which occupations count as AI-exposed (Professionals, Clerical and Administrative Workers, Sales Workers, under the Australian and New Zealand Standard Classification of Occupations (ANZSCO)) is a single judgement call, not a published, validated AI-exposure index, and it has not been cross-checked against a second published exposure measure.
AI-exposed occupations concentrate in wealthy, high-capacity metro councils, the same councils that already score well on ADRI. This component pulls in the opposite direction from the rest of the index, and is a large part of why the overall correlation against ADRI moved further from zero once it was added.
Every staff and finance figure in this index is from the Queensland Government's 2015-16 comparative information release, the most recent year published as open data. More recent years exist only as PDF documents that block automated fetching. Each column carries its own vintage field, so the age of the number travels with it.
The source ADRI data is published at Statistical Area Level 2 (SA2) and aggregated to Local Government Area (LGA) by area times share, because the population fields in the public Application Programming Interface (API) response are null. Area weighting over-weights large, empty SA2s. Joining Australian Bureau of Statistics (ABS) population estimates by SA2 code and swapping the weight would be more accurate, and was not done in the time available.
The 22/18/25/15/10/10 split is reasoned judgement, not a statistical fit to data or an external panel's ranking. The interactive map's sliders let a reader test whether the ranking survives a different weighting.
A single vendor outage affecting many councils at once has no natural hazard analogue and would be one of the cleanest sources of genuine divergence from ADRI. The one data source tried for it, a specific contract disclosure field, had no variance across all 78 councils and could not be scored. Measuring it properly would mean manually reading each council's published contract register, roughly a day of work, judged not worth it against the time available for this hackathon.
47 of 78 council websites had been checked for a published AI governance policy as of this submission. The remaining 31 are disproportionately small, remote councils, which are also the least likely to have an indexed policy to find. Diminishing returns, not an oversight.