{"agency":"Signal (demo · South Australia Police context)","accountable_official":"Accountable Official (demo)","policy":"Policy for the responsible use of AI in government (DTA, v2.0)","ai_systems":["deepseek-v4-pro","signal-stats-v1 (deterministic)"],"purposes":["Crime trend analysis","Multi-step crime trend analysis"],"data_sources":["SA Police Crime statistics (data.sa.gov.au) — FY2024-25 + FY2025-26 YTD — bundled snapshot fetched 2026-06-13 (live refresh in progress)","SA Police Crime statistics (data.sa.gov.au) — FY2024-25 + FY2025-26 YTD, live as of 2026-03"],"risk_tiers":{"limited":6},"human_oversight":"Outputs are reviewable by a human. Statistically anomalous results are flagged for human review before action, and reviews/overrides are recorded against the decision. 6 of 6 logged decisions were flagged for human review.","fairness":"Outputs are aggregate counts, not rates. Differences between regions can reflect population size, reporting behaviour and policing intensity as much as actual offending, so the figures are not used to rank or target places, communities or individuals; comparisons carry this caveat explicitly.","public_access":"Every AI-assisted answer carries a decision_id, and the full audit trail is a public endpoint (GET /decisions). The use-case register and this statement are generated live from that log.","statement":"# AI transparency statement\n\n**Agency:** Signal (demo · South Australia Police context)\n\n**Accountable official:** Accountable Official (demo)\n\n**Policy basis:** Policy for the responsible use of AI in government (DTA, v2.0); mandatory requirements from 15 June 2026.\n\n## What AI we use\ndeepseek-v4-pro, signal-stats-v1 (deterministic).\n\n## What we use it for\nCrime trend analysis, Multi-step crime trend analysis.\n\n## What data informs it\nSA Police Crime statistics (data.sa.gov.au) — FY2024-25 + FY2025-26 YTD — bundled snapshot fetched 2026-06-13 (live refresh in progress), SA Police Crime statistics (data.sa.gov.au) — FY2024-25 + FY2025-26 YTD, live as of 2026-03.\n\n## Risk classification (EU AI Act tiers)\nlimited: 6.\n\n## Human oversight\nOutputs are reviewable by a human. Statistically anomalous results are flagged for human review before action, and reviews/overrides are recorded against the decision. 6 of 6 logged decisions were flagged for human review.\n\n## Fairness\nOutputs are aggregate counts, not rates. Differences between regions can reflect population size, reporting behaviour and policing intensity as much as actual offending, so the figures are not used to rank or target places, communities or individuals; comparisons carry this caveat explicitly.\n\n## Public access and accountability\nEvery AI-assisted answer carries a decision_id, and the full audit trail is a public endpoint (GET /decisions). The use-case register and this statement are generated live from that log.\n"}