Top AI risks mapped to each use case — with controls, residual ratings, owners and review dates.
Staff pasting resident-identifying information into personal ChatGPT accounts. Data may be used for model training, retained indefinitely, breach NSW PPIPA.
Predictive maintenance model on water mains flags pipes for replacement that don't actually need it. Direct financial impact and resident disruption from unnecessary digs.
Chatbot answers planning enquiries from residents. Risk that hallucinated or out-of-date answers lead to non-compliant works, complaints, or Council liability.
Model trained on historical valuations may encode geographic or demographic bias, flagging properties in certain suburbs disproportionately.
LLM summary of development application omits a clause that an assessing officer would have caught. Resulting determination may be challengeable.
Chatbot conversations, AI-drafted documents, and summariser outputs may constitute State Records that aren't being captured.
Existing software vendor (e.g., M365, SirsiDynix) ships AI capability in an update. Council uses it without going through AIRC.
Despite design choice not to retain faces / plates, ML pipeline may briefly process them. Risk of PPIPA breach if data is logged or exported.
Roadside ML, waste optimisation, library recommender all depend on commercial vendors. Vendor exit could disrupt operations or trigger expensive migration.
Resident-facing AI (chatbot, transparency register) produces a high-profile incorrect or offensive output. Media coverage and trust loss.