AIBD / fructus.json
petervartanian's picture
Edition 2026-11-16-bb8aa6fb86ef
00fb693 verified
Raw History Blame Contribute Delete
81.2 kB
[{"id":"DIV-0001","case_id":"AIBD-0168","title":"Xcel customer savings from better wind forecasts","summary":"NCAR's forecasters report Xcel Energy's estimate that improvements in forecasting had saved its customers US$49 million by October 2015.","monetary_role":"outcome_amount_claim","amount":{"source_text":"US $49 million","value":"49000000","lower":null,"upper":null,"qualifier":"exact","currency":"USD","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"cash_saving","amount_basis":"avoided_expenditure","period":"Cumulative to October 2015, when the article appeared. The article gives no start date for the estimate, and Xcel began using an early version of the system in 2009.","unit":"Cumulative savings to Xcel Energy customers.","beneficiary":"Xcel Energy customers.","payer":"Not applicable to avoided costs.","realization":"retrospective_estimate","accounting_boundary":"Xcel's estimate of gross savings from better forecasts. The article gives no calculation and deducts no system or running costs.","comparator":"Forecast error at a typical wind farm in Colorado fell from about 18 percent in 2008 to 10.8 percent by 2014. No monetary baseline is given.","ai_attribution":"DICast corrects the biases of the weather models and blends them, and it keeps adjusting its corrections and weights as it checks its forecasts against outcomes. Xcel's estimate covers forecasting improvements as a whole.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://spectrum.ieee.org/taming-wind-power-with-better-forecasts","title":"Taming Wind Power With Better Forecasts","quote":"Xcel estimates that improvements in forecasting have saved customers US $49 million so far.","date":"2015-10-22","primary":true,"written_by":"makers","reviewed_at":"2026-11-01","review_method":"full_text","locator":"Results paragraph on Xcel's Public Service Company of Colorado"}],"limitations":["The article gives neither Xcel's method nor the start date of the estimate.","The estimate covers the whole forecasting system, so the share of DICast is unknown."],"overlap_flags":["The $6 million that UCAR reported for 2010 and the $10 million a year that E&E News reported in 2017 are separate statements. Neither should be added to this amount."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The IEEE Spectrum article by two NCAR forecasters gives Xcel's estimate of US$49 million saved for customers so far. The amount is curated as a cumulative retrospective estimate for the forecasting system that includes DICast.","reviewed_on":"2026-11-01","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"cash_saving","money_subtype":"electricity balancing and forecast-related operating costs","claim_status":"retrospective_estimate","currency":"USD","currency_basis":"explicit","unit_basis":"cumulative savings across Xcel customers","time_basis":"Cumulative to 2015; the account places forecasting work from 2009, but the amount-specific start is unstated.","time_basis_completeness":"incomplete_window","monetary_comparator_assessment":"Operational forecast-error baseline; no comparable monetary baseline is stated.","ai_and_value_scope":"The estimate concerns forecasting improvements as a whole; the incremental AI share is not established.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Appears to be gross avoided operating cost; implementation and running costs are unstated.","net_benefit_established":false,"comparability_flags":["cumulative_not_annual","whole_program_attribution","ai_component_needs_review","operator_estimate"],"data_gaps":["Amount-specific start date","Calculation method and baseline operating cost","Implementation and ongoing costs","AI component and its incremental contribution"],"extraction_note":"Do not add the separate 2010 saving or prospective annual 2017 figure to the cumulative amount.","source_review_required":true,"id":"AIBD-0168","title":"NCAR's DICast wind forecasts cut Xcel Energy's forecast error to 10.8%; Xcel estimates US$49 million saved","value":{"amount":"US $49 million","what":"saved through forecasting improvements, cumulative to 2015","who":"Xcel Energy customers","estimate":true,"quote":"Xcel estimates that improvements in forecasting have saved customers US $49 million so far.","url":"https://spectrum.ieee.org/taming-wind-power-with-better-forecasts"},"sourceURL":"https://spectrum.ieee.org/taming-wind-power-with-better-forecasts","date_observed_as_stored":"2015-10-22","existing_estimate_flag":true,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"high","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0003","case_id":"AIBD-0196","title":"Kiribati's fine against the owner of Marshalls 203","summary":"An Oceana board member writes that Global Fishing Watch data helped Kiribati secure a $1 million fine against the owner of a vessel that fished illegally in a protected area.","monetary_role":"recipient_transfer","amount":{"source_text":"$1 million","value":"1000000","lower":null,"upper":null,"qualifier":"exact","currency":"USD","currency_basis":"context","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"recipient_transfer","amount_basis":"transfer","period":"Reported on 15 September 2016. The source gives no date for the fine.","unit":"One fine against the Central Pacific Fishing Company, owner of Marshalls 203.","beneficiary":"Government of Kiribati.","payer":"Central Pacific Fishing Company, the vessel's owner.","realization":"awarded","accounting_boundary":"Gross fine secured by the government. The source does not confirm payment, and enforcement costs are not deducted.","comparator":"No monetary comparison is given.","ai_attribution":"Global Fishing Watch data let Kiribati show that the vessel was fishing in the closed area. The platform detects fishing with machine learning, and Kiribati has not said whether the classifier's output or the raw track carried the case.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://grist.org/science/this-new-digital-tool-could-help-fight-overfishing-around-the-world/","title":"This new digital tool can help fight overfishing around the world","quote":"Global Fishing Watch provided data to the government of Kiribati that allowed it to show that Marshalls 203, a Marshall Islands-flagged vessel, was fishing illegally in the Phoenix Islands Protected Area, which is off-limits to commercial fishing. This helped the country secure a $1 million fine against the Central Pacific Fishing Company, the owner of Marshalls 203.","date":"2016-09-15","primary":true,"written_by":"makers","reviewed_at":"2026-11-01","review_method":"full_text","locator":"Paragraph on Kiribati"}],"limitations":["The author sits on the board of Oceana, which developed Global Fishing Watch with SkyTruth and Google.","The source gives no date or proof of payment for the fine."],"overlap_flags":["The company also agreed to pay a further $1 million as a grant under a goodwill arrangement. That grant is not part of this amount.","A 2018 report of a $2 million fine may combine the fine and the grant."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The Grist article separates a $1 million fine from a further $1 million goodwill grant. Only the fine is kept, as a transfer to Kiribati that the source does not show as paid.","reviewed_on":"2026-11-01","legacy_audit":{"monetary_role":"recipient_transfer","money_type":"other","money_subtype":"enforcement fine and separate goodwill grant","claim_status":"fine_secured_and_grant_agreed","currency":"USD","currency_basis":"currency_inferred_from_context","unit_basis":"one enforcement settlement with two differently described payments","time_basis":"Single case reported by 2016; exact payment date absent.","time_basis_completeness":"undated_one_off","monetary_comparator_assessment":"No monetary comparator.","ai_and_value_scope":"Tracking supported enforcement; classifier contribution and payment realization remain unclear.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross government receipts or commitments; enforcement costs and losses to the payer are not included.","net_benefit_established":false,"comparability_flags":["transfer_not_total_social_value","multiple_payment_components","fine_grant_conflict","ai_increment_unresolved"],"data_gaps":["Proof and dates of payment","Separate fine from grant","Reconcile later reporting that describes a $2 million fine","Counterfactual enforcement outcome"],"extraction_note":"The amount field is $1 million, while what/quote also describe a separate $1 million grant. Do not silently treat the sum as one fine.","source_review_required":true,"id":"AIBD-0196","title":"Global Fishing Watch data lets Kiribati fine a vessel US$1 million for fishing in the Phoenix Islands Protected Area","value":{"amount":"$1 million","what":"fine secured by Kiribati against the owner of Marshalls 203 for fishing illegally in the Phoenix Islands Protected Area, plus an additional $1 million paid as a grant under a goodwill arrangement","who":"Government of Kiribati","estimate":false,"quote":"This helped the country secure a $1 million fine against the Central Pacific Fishing Company, the owner of Marshalls 203. The company also made a special “goodwill arrangement” with the government, agreeing to pay an additional $1 million in the form of a grant.","url":"https://grist.org/science/this-new-digital-tool-could-help-fight-overfishing-around-the-world/"},"sourceURL":"https://grist.org/science/this-new-digital-tool-could-help-fight-overfishing-around-the-world/","date_observed_as_stored":"2016-09-15","existing_estimate_flag":false,"existing_comparator_present":false,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"high","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0005","case_id":"AIBD-0105","title":"Monetized climate value of GLAD alerts","summary":"Researchers value avoided deforestation associated with African GLAD alert subscriptions at US$149–696 million using the social cost of carbon.","monetary_role":"outcome_amount_claim","amount":{"source_text":"US$149–696 million","value":null,"lower":"149000000","upper":"696000000","qualifier":"range","currency":"USD","currency_basis":"explicit","price_basis":"Social-cost-of-carbon valuation; valuation and price year are not specified in the checked abstract."},"additional_amounts":[],"financial_type":"monetized_social_benefit","amount_basis":"valuation","period":"Comparison with pre-2016 deforestation; the checked abstract does not specify the valuation horizon.","unit":"Modeled value of avoided deforestation in the African study sample.","beneficiary":"People benefiting from avoided climate damage; no direct cash recipient.","payer":"Not applicable to a social-cost valuation.","realization":"modeled_valuation","accounting_boundary":"Modeled climate value, not cash income, market revenue, or a fully costed net-benefit estimate.","comparator":"Deforestation associated with subscriptions versus the study counterfactual. Alert availability alone had no significant effect.","ai_attribution":"GLAD’s machine-learning forest-change alerts are inputs to the intervention. The evaluation concerns subscriptions and subsequent responses, not model performance in isolation.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://ageconsearch.umn.edu/record/304666/files/18339%20manuscript%20ABSTRACT%20ONLY.pdf","title":"Near Real-Time Forest Monitoring Can Decrease Deforestation","quote":"we estimate a value of the alert system between USD 149 million and 696 million.","date":null,"primary":true,"written_by":"unknown","reviewed_at":"2026-10-23","review_method":"full_text","locator":"Author abstract, single-page PDF"}],"limitations":["Only the author abstract was checked for this monetary range; the full valuation method remains unreviewed.","The reported deforestation effect was specific to Africa, with no detected effect on other continents."],"overlap_flags":["The interval represents alternative valuation assumptions, not two additive benefits."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"other","money_subtype":"monetized environmental and social benefit","claim_status":"retrospective_modeled_valuation","currency":"USD","currency_basis":"explicit","unit_basis":"avoided deforestation in the African study region","time_basis":"Study includes a 2011-2016 baseline; the valuation window is not explicit in value.","time_basis_completeness":"incomplete_window","monetary_comparator_assessment":"The account describes a counterfactual for deforestation; no cash-flow comparator.","ai_and_value_scope":"Valuation concerns the alert system with human uptake and response, not AI in isolation.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Social-cost-of-carbon valuation; net costs and broader welfare effects are not supplied.","net_benefit_established":false,"comparability_flags":["nonmarket_valuation","range_not_point_estimate","valuation_assumption_sensitive","whole_program_attribution"],"data_gaps":["Valuation period","Carbon quantity and social cost of carbon assumptions","Price year and discount rate","Program costs and uncertainty method"],"extraction_note":"This is a monetary valuation of an environmental outcome, not cash saved or revenue received.","source_review_required":true,"id":"AIBD-0105","title":"GLAD deforestation alert subscriptions reduce deforestation probability in Africa by 18%","value":{"amount":"between US$149 million and US$696 million","what":"value of avoided deforestation in Africa, priced at the social cost of carbon","who":"society, through avoided carbon emissions","estimate":true,"quote":"Calculated using the social cost of carbon for avoided deforestation in Africa, we estimate the alert system’s value to be between US$149 million and US$696 million.","url":"https://www.nature.com/articles/s41558-020-00956-w"},"sourceURL":"https://www.nature.com/articles/s41558-020-00956-w","date_observed_as_stored":"2021-01-04","existing_estimate_flag":true,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0006","case_id":"AIBD-0193","title":"Avoided East African locust losses","summary":"FAO estimates that the locust response avoided $1.56 billion in cereal and milk losses.","monetary_role":"outcome_amount_claim","amount":{"source_text":"US$1.56 billion","value":"1560000000","lower":null,"upper":null,"qualifier":"exact","currency":"USD","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"avoided_loss","amount_basis":"avoided_expenditure","period":"East African response reported May 6, 2021; the amount-specific start date is unstated.","unit":"Cereal and milk losses avoided across the regional response.","beneficiary":"Farmers, livestock keepers, and food consumers in affected East African countries.","payer":"Not applicable to avoided production losses.","realization":"retrospective_estimate","accounting_boundary":"Estimated gross production losses avoided by the response; response costs are not deducted.","comparator":"Estimated losses without the control response; calculation details are not given in this account.","ai_attribution":"PlantVillage AI helped clean monitoring data within a larger response involving field teams, aircraft, forecasts, and pest control. The AI-specific increment is unknown.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://www.fao.org/newsroom/story/A-high-tech-response-is-helping-countries-win-the-battle-against-Desert-Locusts/en","title":"A high-tech response is helping countries win the battle against Desert Locusts","quote":"avoiding USD 1.56 billion in cereal and milk losses.","date":"2021-05-06","primary":true,"written_by":"deployer","reviewed_at":"2026-10-23","review_method":"full_text","locator":null}],"limitations":["FAO reports a whole-response counterfactual estimate, not cash received or an independently isolated AI effect."],"overlap_flags":["The cereal and milk quantities describe the same avoided-loss estimate and are not additional monetary gains."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"avoided_loss","money_subtype":"agricultural production losses","claim_status":"retrospective_counterfactual_estimate","currency":"USD","currency_basis":"explicit","unit_basis":"entire East African locust-control campaign","time_basis":"Campaign beginning in 2020, reported in May 2021; precise valuation cutoff needs checking.","time_basis_completeness":"incomplete_window","monetary_comparator_assessment":"Assumed no-control-operations counterfactual; no quantified monetary baseline in the structured comparator.","ai_and_value_scope":"The amount belongs to the whole control operation, including aircraft and field action, to which AI-supported information contributed.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross avoided production loss; campaign costs are not deducted.","net_benefit_established":false,"comparability_flags":["whole_program_attribution","counterfactual_estimate","agricultural_valuation","not_AI_only"],"data_gaps":["Valuation method and price year","Precise campaign period and geographic coverage","AI contribution","Cost of control operations"],"extraction_note":"The amount must not be described as the app or AI alone generating USD 1.56 billion.","source_review_required":true,"id":"AIBD-0193","title":"eLocust3m and AI data cleaning feed FAO's East Africa locust control, which may have averted USD 1.56 billion in losses","value":{"amount":"USD 1.56 billion","what":"cereal and milk losses avoided by the locust control operations in East Africa that the app and AI data cleaning fed","who":"East African farmers and herders (the food security of 36.6 million people)","estimate":true,"quote":"In East Africa locust control operations have prevented the loss of 4 million tonnes of cereal and 800 million litres of milk production, while protecting the food security of 36.6 million people and avoiding USD 1.56 billion in cereal and milk losses.","url":"https://www.fao.org/newsroom/story/A-high-tech-response-is-helping-countries-win-the-battle-against-Desert-Locusts/en"},"sourceURL":"https://www.fao.org/newsroom/story/A-high-tech-response-is-helping-countries-win-the-battle-against-Desert-Locusts/en","date_observed_as_stored":"2021-05-06","existing_estimate_flag":true,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0010","case_id":"AIBD-0031","title":"Lower household medical spending after flood alerts","summary":"A randomized study estimates that community dissemination of AI flood alerts reduced household illness-treatment spending by roughly INR6,500.","monetary_role":"outcome_amount_claim","amount":{"source_text":"About INR6,500 per household","value":"6500","lower":null,"upper":null,"qualifier":"approximately","currency":"INR","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"cash_saving","amount_basis":"incremental_change","period":"2022–2023 flood-season study; the prose does not isolate the monetary recall window.","unit":"Estimated difference per household in illness-treatment spending.","beneficiary":"Households in treatment communities in Bihar, India.","payer":"Households paying for illness treatment.","realization":"retrospective_estimate","accounting_boundary":"Estimated household medical-expenditure reduction, not a full net-welfare or program-profit estimate.","comparator":"Randomized community dissemination versus control communities, both with access to the AI forecasting system.","ai_attribution":"The randomized intervention is dissemination of AI forecasts, not AI versus no AI. Its effect cannot be relabeled as the forecasting model’s isolated contribution.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://bpb-us-w2.wpmucdn.com/campuspress.yale.edu/dist/7/2986/files/2024/04/Google_Floods-bcdf49042e23059a.pdf","title":"Forecasting Fate: Experimental Evaluation of a Flood Early Warning System","quote":"households in treatment areas incurred roughly INR 6,500 less on illness treatment","date":"2024-04-03","primary":true,"written_by":"independent","reviewed_at":"2026-10-23","review_method":"full_text","locator":"PDF page 6; medical-expenditure result and Table 4"}],"limitations":["The manuscript reports preliminary estimates and no evidence of broader improvements in economic well-being.","The reported amount is a study estimate, not a guaranteed saving for each household."],"overlap_flags":["The paper’s benefit-cost ratio uses these medical savings; do not count it as another benefit."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"cash_saving","money_subtype":"household medical expenditure","claim_status":"study_estimate_of_observed_spending_difference","currency":"INR","currency_basis":"explicit","unit_basis":"per household over a flood season","time_basis":"2022 and 2023 monsoon study; the amount-specific season requires confirmation.","time_basis_completeness":"partial_window","monetary_comparator_assessment":"Randomized comparison for community dissemination; control households already had access to AI phone alerts.","ai_and_value_scope":"Incremental effect of community dissemination added to forecast access, not AI forecasts versus no AI.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Household expenditure saving before program costs; a separate benefit-cost ratio does not make the headline amount net.","net_benefit_established":false,"comparability_flags":["per_household","study_estimate_despite_false_flag","comparator_already_has_AI","season_specific","increment_is_delivery_layer"],"data_gaps":["Which season and sample support INR 6,500","Confidence interval and attrition","How illness spending was measured","Program cost and ratio assumptions"],"extraction_note":"A measured study estimate still has sampling uncertainty; estimate=false should not imply an exact cash-accounting total.","source_review_required":true,"id":"AIBD-0031","title":"Randomized evaluation of Google AI flood alerts with community dissemination in Bihar, India (2022–2023 monsoons)","value":{"amount":"INR 6,500","what":"less spent on illness treatment per household over the flood season","who":"treatment households in the Bihar study panchayats","estimate":false,"quote":"As a result, households in treatment areas incurred roughly INR 6,500 less on illness treatment (a 32.87% decrease). In fact, based on the estimated reduction in medical expenditures alone, we estimate that for every $1 spent on the program, the intervention would generate a minimum benefit of $30 during a severe flooding season.","url":"https://bpb-us-w2.wpmucdn.com/campuspress.yale.edu/dist/7/2986/files/2024/04/Google_Floods-bcdf49042e23059a.pdf"},"sourceURL":"https://bpb-us-w2.wpmucdn.com/campuspress.yale.edu/dist/7/2986/files/2024/04/Google_Floods-bcdf49042e23059a.pdf","date_observed_as_stored":"2022-06-01","existing_estimate_flag":false,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0012","case_id":"AIBD-0116","title":"Parking fine withdrawn after a ChatGPT-drafted appeal","summary":"The BBC reports that York City Council withdrew a £60 parking fine after a student appealed with a letter that ChatGPT drafted.","monetary_role":"outcome_amount_claim","amount":{"source_text":"£60","value":"60","lower":null,"upper":null,"qualifier":"exact","currency":"GBP","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"cash_saving","amount_basis":"avoided_expenditure","period":"Reported on 31 March 2023. Neither the date of the fine nor that of its withdrawal is given.","unit":"One parking fine.","beneficiary":"A student at York St John University.","payer":"Not applicable to a withdrawn fine.","realization":"retrospective_estimate","accounting_boundary":"Personal payment avoided. The council forgoes the same £60.","comparator":"The student says that she was tempted to pay the fine rather than spend time on a response.","ai_attribution":"ChatGPT drafted the appeal from details that the student entered. She says that the fine was wrongly issued, as she held a permit for her street.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://www.bbc.co.uk/news/uk-england-york-north-yorkshire-65126772","title":"York student uses AI chatbot to get parking fine revoked","quote":"A student has successfully appealed against a £60 parking fine by using a letter written by an artificial intelligence chatbot.","date":"2023-03-31","primary":true,"written_by":"independent","reviewed_at":"2026-11-01","review_method":"full_text","locator":"Opening paragraph"}],"limitations":["The council has not published its reasons for withdrawing the fine.","Her appeal rested on a valid permit, which she could also have cited in a letter of her own."],"overlap_flags":["The council forgoes the same £60, so the amount is a transfer between the student and the council."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The BBC report says that the council withdrew a £60 parking fine after an appeal letter that ChatGPT drafted. The amount is curated as the student's avoided payment, which rests on her account that she would have paid.","reviewed_on":"2026-11-01","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"cash_saving","money_subtype":"personal liability or fine avoided","claim_status":"reported_liability_withdrawal","currency":"GBP","currency_basis":"symbol_and_UK_context","unit_basis":"one parking fine","time_basis":"Case reported March 2023; exact withdrawal date absent.","time_basis_completeness":"undated_one_off","monetary_comparator_assessment":"Narrative baseline is the student paying the fine; no formal comparative design.","ai_and_value_scope":"AI drafted the appeal; council decision depended on a valid permit and other facts.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross personal payment avoided; transfer effect for the council and appeal costs are separate.","net_benefit_established":false,"comparability_flags":["beneficiary_specific_transfer","single_case","counterfactual_self_report"],"data_gaps":["Council decision and date","AI contribution relative to entitlement","Time and other appeal costs"],"extraction_note":"This is a concrete personal financial benefit; it is not automatically a net social gain of the same amount.","source_review_required":true,"id":"AIBD-0116","title":"York student has a £60 parking fine withdrawn after sending an appeal letter drafted by ChatGPT","value":{"amount":"£60","what":"parking fine withdrawn by York City Council after the ChatGPT-drafted appeal","who":"A student at York St John University","estimate":false,"quote":"A student has successfully appealed against a £60 parking fine by using a letter written by an artificial intelligence chatbot.","url":"https://www.bbc.co.uk/news/uk-england-york-north-yorkshire-65126772"},"sourceURL":"https://www.bbc.co.uk/news/uk-england-york-north-yorkshire-65126772","date_observed_as_stored":"2023-03-31","existing_estimate_flag":false,"existing_comparator_present":false,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0014","case_id":"AIBD-0013","title":"Vesuvius Challenge grand prize","summary":"The competition awarded $700,000 to the three-person team that used machine learning to read a carbonized scroll.","monetary_role":"recipient_transfer","amount":{"source_text":"US$700,000","value":"700000","lower":null,"upper":null,"qualifier":"exact","currency":"USD","currency_basis":"context","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"recipient_transfer","amount_basis":"transfer","period":"Grand-prize announcement on February 5, 2024.","unit":"One prize shared by a three-person team.","beneficiary":"Youssef Nader, Luke Farritor, and Julian Schilliger.","payer":"Vesuvius Challenge prize fund.","realization":"awarded","accounting_boundary":"Gross prize award to recipients, not new aggregate wealth or research profit.","comparator":"No cash counterfactual is stated.","ai_attribution":"Machine learning supported the winning scroll-reading work. The award is a financial consequence of that work, without an isolated AI share.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://scrollprize.org/grandprize","title":"Vesuvius Challenge 2023 Grand Prize awarded","quote":"the Vesuvius Challenge Grand Prize of $700,000 is awarded to a team of three","date":"2024-02-05","primary":true,"written_by":"deployer","reviewed_at":"2026-10-23","review_method":"full_text","locator":null}],"limitations":["The announcement establishes an award, not completed payment or the winners’ net proceeds."],"overlap_flags":["One shared prize, not $700,000 per team member."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"recipient_transfer","money_type":"other","money_subtype":"prize award to winning researchers","claim_status":"awarded","currency":"USD","currency_basis":"inferred_from_competition_context","unit_basis":"one team award shared among three winners","time_basis":"Award announced February 5, 2024","time_basis_completeness":"dated_one_off","monetary_comparator_assessment":"The comparator measures scroll text recovered, not financial gain.","ai_and_value_scope":"Prize awarded for AI-assisted work; award is not a valuation of recovered knowledge.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross award to recipients; research costs and payer outlay are separate.","net_benefit_established":false,"comparability_flags":["recipient_transfer","prize_not_research_value","gross_award"],"data_gaps":["Payment confirmation and recipient split if relevant","Research costs","Keep award distinct from monetized cultural value"],"extraction_note":"Potentially useful in a recipient-benefit or research-incentive view, but not a quantified improvement in the cost or value of reading the scroll.","source_review_required":true,"id":"AIBD-0013","title":"Vesuvius Challenge 2023: machine-learning ink detection reads over 2,000 characters from unopened Herculaneum scroll PHerc.Paris.4","value":{"amount":"$700,000","what":"Vesuvius Challenge grand prize awarded for the winning submission that read the scroll","who":"Youssef Nader, Luke Farritor and Julian Schilliger, paid by the Vesuvius Challenge","estimate":false,"quote":"The results of this review were clear and unanimous: the Vesuvius Challenge Grand Prize of $700,000 is awarded to a team of three for their excellent submission. Congratulations to Youssef Nader, Luke Farritor, and Julian Schilliger!","url":"https://scrollprize.org/grandprize"},"sourceURL":"https://scrollprize.org/grandprize","date_observed_as_stored":"2024-02-05","existing_estimate_flag":false,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":true,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0017","case_id":"AIBD-0095","title":"Flood-triggered cash transfers in Nigeria","summary":"GiveDirectly reports that it paid $105 per person in an AI-informed flood-response pilot in Kogi, Nigeria.","monetary_role":"recipient_transfer","amount":{"source_text":"US$105 per recipient","value":"105","lower":null,"upper":null,"qualifier":"exact","currency":"USD","currency_basis":"context","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[{"id":"ngn-equivalent","amount":{"source_text":"NGN177,866 per recipient","value":"177866","lower":null,"upper":null,"qualifier":"exact","currency":"NGN","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"relationship":"source_equivalent","equivalent_to":"primary","equivalence_note":"Reported equivalent; later naira payments may differ.","period":"August 29 to October 24, 2024, for the first transfer.","unit":"Per recipient, first transfer.","realization":"paid","accounting_boundary":"Recipient cash transfer. A transfer is not a net increase in societal resources or an estimate of avoided flood loss.","sources":[{"url":"https://www.aljazeera.com/features/2024/9/21/could-ai-save-nigerians-from-devastating-floods","title":"Could AI save Nigerians from devastating floods?","quote":"177,866 naira ($105)","date":"2024-09-21","primary":false,"written_by":"independent","reviewed_at":"2026-10-23","review_method":"full_text","locator":"First anticipatory payment in Kogi"}],"source_review_status":"source_checked","review_note":"Denomination checked."}],"financial_type":"recipient_transfer","amount_basis":"transfer","period":"August 29 to October 24, 2024, for the first transfer.","unit":"Per recipient, first transfer.","beneficiary":"Registered recipients in six wards of Kogi, Nigeria.","payer":"GiveDirectly and its funders.","realization":"paid","accounting_boundary":"Recipient cash transfer. A transfer is not a net increase in societal resources or an estimate of avoided flood loss.","comparator":"The amount is not a comparison with aid delivered without AI.","ai_attribution":"AI forecasts informed targeting and payment triggers alongside local observations and human verification.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://www.givedirectly.org/flood-forecast-ai","title":"How AI helped send cash just before floods hit in Nigeria","quote":"we sent $105 to over 4,600 individuals","date":"2025-06-27","primary":true,"written_by":"deployer","reviewed_at":"2026-10-23","review_method":"full_text","locator":null}],"limitations":["The detailed account says that 3,250 people received early payments and another 1,400 were paid after the October 21 peak; the summary overstates early coverage.","Transfer amount is implementer-reported."],"overlap_flags":["A later $210 transfer is separate. Do not treat the $105 amount as a program total."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"recipient_transfer","money_type":"other","money_subtype":"anticipatory cash transfer","claim_status":"reported_paid_transfer_with_timing_dispute","currency":"USD","currency_basis":"currency_inferred_from_implementer_context","unit_basis":"per recipient; the account reports ambiguity between individual and household","time_basis":"Transfers August-October 2024; a separate $210 transfer followed.","time_basis_completeness":"explicit_window","monetary_comparator_assessment":"Income and food-security before/after survey; no controlled monetary-impact estimate.","ai_and_value_scope":"AI contributed to targeting and timing; the cash is the program transfer, not the incremental value of AI.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross recipient transfer and donor outlay; not net social gain.","net_benefit_established":false,"comparability_flags":["recipient_transfer","intervention_input","recipient_denominator_ambiguous","late_payment_subset","no_control"],"data_gaps":["Individual versus household denominator","Which payments preceded flooding","Amount-specific cohort","AI timing benefit relative to unconditional or later cash","Program costs"],"extraction_note":"The $105 can be tracked as money received, while the separate income change is a distinct outcome claim.","source_review_required":true,"id":"AIBD-0095","title":"GiveDirectly paid anticipatory cash in Kogi State, Nigeria, on Google Flood Hub forecast triggers (August to October 2024)","value":{"amount":"$105","what":"anticipatory cash transfer per recipient before the flood peak (a second $210 transfer followed)","who":"over 4,600 flood-exposed residents of Ibaji, Kogi State","estimate":false,"quote":"Once forecasted flood thresholds were met, we sent $105 to over 4,600 individuals before the flood peak.","url":"https://www.givedirectly.org/flood-forecast-ai/"},"sourceURL":"https://www.givedirectly.org/flood-forecast-ai/","date_observed_as_stored":"2024-08-29","existing_estimate_flag":false,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":true,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0018","case_id":"AIBD-0144","title":"Blackpool road-maintenance savings","summary":"Blackpool Council reports that technology-assisted repairs cost over £1 million less than its estimate for traditional methods.","monetary_role":"outcome_amount_claim","amount":{"source_text":"Over £1 million","value":"1000000","lower":null,"upper":null,"qualifier":"more_than","currency":"GBP","currency_basis":"context","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"cash_saving","amount_basis":"incremental_change","period":"Since summer 2019 to the council account; the endpoint is not specified in the text.","unit":"5,145 potholes in Blackpool’s North Shore.","beneficiary":"Blackpool Council.","payer":"Council road-maintenance budget; savings are avoided expenditure.","realization":"retrospective_estimate","accounting_boundary":"Reported repair-cost difference against an estimated traditional-method cost, without a fully costed AI-only comparison.","comparator":"About £1.5 million under traditional methods versus just under £450,000 reported cost.","ai_attribution":"AI-assisted condition surveys supported a wider maintenance approach. The source does not isolate AI from the repair and workflow changes.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://blackpoolunlimited.com/artificial-intelligence-boosts-road-maintenance-efficiency/","title":"Artificial intelligence boosts road maintenance efficiency","quote":"save over £1m resulting in a final cost of just under £450k.","date":null,"primary":true,"written_by":"deployer","reviewed_at":"2026-10-23","review_method":"full_text","locator":"Council account of Project Amber’s North Shore repairs"}],"limitations":["Council-reported costs and a modeled traditional-method baseline."],"overlap_flags":["The two underlying cost figures form this difference and must not be added as separate benefits."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"cash_saving","money_subtype":"road-repair expenditure avoided","claim_status":"reported_saving_against_estimated_traditional_cost","currency":"GBP","currency_basis":"symbol_and_UK_context","unit_basis":"5,145 potholes in North Shore","time_basis":"Since summer 2019; exact end of repair cohort not stated on the cited value page.","time_basis_completeness":"incomplete_window","monetary_comparator_assessment":"Value/account compare about £1.5m traditional cost with under £450k actual cost; the structured comparator instead concerns compensation payouts.","ai_and_value_scope":"The entire Project Amber process combines imagery, prioritization and surface treatment.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross repair saving relative to a traditional-cost estimate; total technology/program costs are unspecified.","net_benefit_established":false,"comparability_flags":["whole_program_attribution","comparator_field_different_outcome","estimated_counterfactual","cohort_end_unclear"],"data_gaps":["Repair-cohort period","Basis of traditional-cost estimate","Like-for-like repair quality and scope","All implementation and ongoing costs"],"extraction_note":"Do not combine repair savings with reduced compensation payouts without distinct claims and overlap checks.","source_review_required":true,"id":"AIBD-0144","title":"Blackpool Council's Project Amber imagery and AI road surveys: 5,145 potholes filled for about £450,000 and pothole claim payouts down to £719","value":{"amount":"over £1m","what":"saved on filling 5,145 potholes in North Shore, final cost just under £450k against £1.5m by traditional methods","who":"Blackpool Council","estimate":false,"quote":"Blackpool Council has managed to save over £1m resulting in a final cost of just under £450k.","url":"https://blackpoolunlimited.com/artificial-intelligence-boosts-road-maintenance-efficiency/"},"sourceURL":"https://blackpoolunlimited.com/artificial-intelligence-boosts-road-maintenance-efficiency/","date_observed_as_stored":"2024-09-26","existing_estimate_flag":false,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0019","case_id":"AIBD-0173","title":"Treasury check-fraud recoveries","summary":"Treasury reports that machine learning helped identify check fraud associated with $1 billion recovered in fiscal 2024.","monetary_role":"outcome_amount_claim","amount":{"source_text":"US$1 billion","value":"1000000000","lower":null,"upper":null,"qualifier":"exact","currency":"USD","currency_basis":"context","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"recovered_funds","amount_basis":"incremental_change","period":"U.S. fiscal year 2024.","unit":"Treasury check-fraud recoveries across the program.","beneficiary":"U.S. Treasury and affected federal programs.","payer":"Recovered fraud proceeds; individual payers are not identified.","realization":"reported_recovery","accounting_boundary":"Gross recovered funds. Detection, recovery, and operating costs are not deducted.","comparator":"The release does not estimate recoveries without machine learning.","ai_attribution":"Treasury states that machine learning expedited check-fraud identification. The amount is not an estimate of the incremental gain from AI alone.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://home.treasury.gov/news/press-releases/jy2650","title":"Treasury announces enhanced fraud detection processes, including machine learning AI","quote":"Expediting the identification of Treasury check fraud with machine learning AI resulting in $1 billion in recovery.","date":"2024-10-17","primary":true,"written_by":"deployer","reviewed_at":"2026-10-23","review_method":"full_text","locator":null}],"limitations":["Agency-reported recovery, without a counterfactual AI-only estimate or full program costs."],"overlap_flags":["Component of Treasury’s broader fiscal-2024 recovery and prevention total; do not add it to that total."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"avoided_loss","money_subtype":"recovered check fraud","claim_status":"reported_recovery","currency":"USD","currency_basis":"inferred_from_US_program","unit_basis":"annual machine-learning check-fraud recovery","time_basis":"FY2024: October 2023-September 2024","time_basis_completeness":"explicit_window","monetary_comparator_assessment":"Structured comparator compares the full prevention/recovery program across years, not the $1bn AI component.","ai_and_value_scope":"Treasury explicitly attributes this component to machine-learning identification; incremental causality remains unevaluated.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross recovered funds; recovery and system costs unspecified.","net_benefit_established":false,"comparability_flags":["component_vs_program_comparator","recovered_not_prevented","operator_claim","overlap_with_4_billion_total"],"data_gaps":["Recovery accounting and audit","Program costs","Equivalent prior-year ML recovery","Incremental contribution relative to investigators"],"extraction_note":"The $1bn is a component of the greater-than-$4bn total, not an amount to add to that total.","source_review_required":true,"id":"AIBD-0173","title":"Treasury's machine-learning check-fraud detection recovers $1 billion in FY2024","value":{"amount":"$1 billion","what":"Treasury check fraud recovered in FY2024 through machine-learning identification","who":"U.S. Treasury, on behalf of taxpayers","estimate":false,"quote":"Expediting the identification of Treasury check fraud with machine learning AI resulting in $1 billion in recovery.","url":"https://home.treasury.gov/news/press-releases/jy2650"},"sourceURL":"https://home.treasury.gov/news/press-releases/jy2650","date_observed_as_stored":"2024-10-17","existing_estimate_flag":false,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":true,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0022","case_id":"AIBD-0081","title":"Re-estimate of a nursing-supplies tender at Hospital Federal de Ipanema","summary":"The CGU reports that, after an ALICE alert and an audit, Hospital Federal de Ipanema cut the estimated value of tender 90017/2024 by R$3,063,774.84.","monetary_role":"outcome_amount_claim","amount":{"source_text":"R$ 3.063.774,84","value":"3063774.84","lower":null,"upper":null,"qualifier":"exact","currency":"BRL","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"cash_saving","amount_basis":"avoided_expenditure","period":"Tender of 2024, reported on 14 January 2025. The date of the re-estimate is not given.","unit":"Cut in the estimated value of tender 90017/2024 for nursing supplies.","beneficiary":"Hospital Federal de Ipanema and the federal budget.","payer":"Not applicable to avoided spending.","realization":"retrospective_estimate","accounting_boundary":"Reference-price cost avoidance. The amount measures a cut to the estimated purchase value, not savings on prices finally paid.","comparator":"Original estimate of R$15,450,714.80 against a revised estimate of R$12,386,939.96.","ai_attribution":"ALICE raised alerts on the tender, and auditors confirmed an overpricing risk of R$5,994,645.60. Hospital managers then made the new estimate.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://web.archive.org/web/20250114172015id_/https://www.gov.br/cgu/pt-br/assuntos/noticias/2024/12/atuacao-do-robo-alice-segue-auxiliando-auditores-a-realizar-trabalhos-preventivos-que-ajudam-a-evitar-desperdicios-e-irregularidades/","title":"Atuação do Robô ALICE segue auxiliando auditores a realizar trabalhos preventivos, que ajudam a evitar desperdícios e irregularidades","quote":"Como resultado da atuação, os gestores procederam nova estimativa de preços, reduzindo o valor total estimado da compra para R$ 12.386.939,96, o que representou uma economia de mais de R$ 3 milhões (R$ 3.063.774,84).","date":"2025-01-14","primary":true,"written_by":"makers","reviewed_at":"2026-11-01","review_method":"full_text","locator":"Section on Hospital Federal de Ipanema, tender 90017/2024 (Relatório 1686781). Archived copy of the CGU news page."}],"limitations":["Final purchase prices under the tender are not reported."],"overlap_flags":["The overpricing risks flagged in the three tenders are not savings and must not be added.","CGU's yearly ALICE savings totals may include this re-estimate."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The CGU release gives the cut in the estimated value of tender 90017/2024 as R$3,063,774.84, from R$15,450,714.80 to R$12,386,939.96. It is curated as reference-price cost avoidance for this one tender.","reviewed_on":"2026-11-01","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"cash_saving","money_subtype":"procurement estimate reduction","claim_status":"reported_budget_reduction","currency":"BRL","currency_basis":"explicit","unit_basis":"tender 90017/2024 at Hospital Federal de Ipanema","time_basis":"2024 tender; result reported January 14, 2025; precise revision date absent.","time_basis_completeness":"partial_window","monetary_comparator_assessment":"Explicit original R$15,450,714.80 versus revised R$12,386,939.96.","ai_and_value_scope":"ALICE alert followed by auditors and hospital managers.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross reduction of an estimate; actual paid saving and audit/system costs unknown.","net_benefit_established":false,"comparability_flags":["budget_revision_not_cash_release","single_component_of_three_cases","possible_overlap_with_program_aggregates"],"data_gaps":["Final contract and payment","Equivalent quantity and quality","Revision date","Avoid overlap with broader audits or annual totals"],"extraction_note":"The cited R$3,063,774.84 is a documented estimate change, not the sum of all three tenders or all flagged overpricing risks.","source_review_required":true,"id":"AIBD-0081","title":"CGU's ALICE robot flags overpricing in three 2024 Rio de Janeiro federal hospital tenders; one suspended, one revoked, one re-estimated","value":{"amount":"R$ 3.063.774,84","what":"reduction of the estimated value of tender 90017/2024 after the re-estimate (the release gives no total for the three tenders)","who":"Hospital Federal de Ipanema, federal budget","estimate":false,"quote":"Como resultado da atuação, os gestores procederam nova estimativa de preços, reduzindo o valor total estimado da compra para R$ 12.386.939,96, o que representou uma economia de mais de R$ 3 milhões (R$ 3.063.774,84).","url":"https://web.archive.org/web/20250114172015id_/https://www.gov.br/cgu/pt-br/assuntos/noticias/2024/12/atuacao-do-robo-alice-segue-auxiliando-auditores-a-realizar-trabalhos-preventivos-que-ajudam-a-evitar-desperdicios-e-irregularidades/"},"sourceURL":"https://web.archive.org/web/20250114172015id_/https://www.gov.br/cgu/pt-br/assuntos/noticias/2024/12/atuacao-do-robo-alice-segue-auxiliando-auditores-a-realizar-trabalhos-preventivos-que-ajudam-a-evitar-desperdicios-e-irregularidades/","date_observed_as_stored":"2025-01-14","existing_estimate_flag":false,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0023","case_id":"AIBD-0126","title":"Cash recovered after a Chandni Chowk robbery","summary":"Delhi Police say that facial recognition helped identify a robbery suspect and that a raid then recovered Rs 79.5 lakh of the Rs 80 lakh taken.","monetary_role":"outcome_amount_claim","amount":{"source_text":"Rs 79.5 lakh","value":"7950000","lower":null,"upper":null,"qualifier":"exact","currency":"INR","currency_basis":"context","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"recovered_funds","amount_basis":"incremental_change","period":"Robbery on 17 March 2025 and recovery reported on 19 March 2025.","unit":"Cash recovered in one raid after one robbery.","beneficiary":"R K Enterprises, whose courier was robbed.","payer":"No payer. Police seized the cash in a raid on the home of one of the two arrested men.","realization":"reported_recovery","accounting_boundary":"Gross cash seized by police. Its return to the owner and the investigation costs are not reported.","comparator":"Rs 80 lakh was taken and Rs 79.5 lakh was recovered. No comparison without facial recognition is given.","ai_attribution":"Facial recognition matched a camera frame of the first suspect to a prior arrest record with a similarity score of 32.09 percent. Questioning and a raid then led to the cash.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://www.rediff.com/news/report/israeli-tech-helps-delhi-police-crack-rs-80-lakh-robbery-2-held/20250319.htm","title":"Israeli tech helps Delhi police crack Rs 80 lakh robbery; 2 held","quote":"led to the recovery of Rs 79.5 lakh in cash, a semi-automatic pistol with three bullets, and other incriminating materials","date":"2025-03-19","primary":true,"written_by":"independent","reviewed_at":"2026-11-01","review_method":"full_text","locator":"PTI report that quotes the deputy commissioner of police (North)"}],"limitations":["The officials' summary speaks of the entire Rs 80 lakh, while their account of the raid gives Rs 79.5 lakh.","The software and its vendor are not named."],"overlap_flags":["The Rs 80 lakh stolen is the loss and is not a second benefit."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The PTI report quotes police as saying that a raid after a facial-recognition lead recovered Rs 79.5 lakh in cash. The amount is curated as recovered funds, with the face match as one step in the investigation.","reviewed_on":"2026-11-01","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"avoided_loss","money_subtype":"recovered stolen cash","claim_status":"reported_recovery","currency":"INR","currency_basis":"inferred_from_Indian_rupees_context","unit_basis":"one robbery and cash seizure","time_basis":"Robbery March 17, 2025; recovery by March 19, 2025","time_basis_completeness":"dated_one_off","monetary_comparator_assessment":"Rs 80 lakh stolen and Rs 79.5 lakh recovered provide a recovery denominator; there is no no-AI counterfactual.","ai_and_value_scope":"Face matching was one investigative lead; conventional police work produced recovery.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross cash recovered; return to the victim and investigation costs are not documented.","net_benefit_established":false,"comparability_flags":["recovery_not_expected_loss","human_investigation_chain","restoration_not_new_production"],"data_gaps":["Cash returned to owner versus held as evidence","Investigative costs","AI incremental contribution"],"extraction_note":"Rs 79.5 lakh is the recovery amount; Rs 80 lakh is the robbery total, not another benefit.","source_review_required":true,"id":"AIBD-0126","title":"Delhi Police matched a CCTV frame to a prior arrest record with Israeli facial recognition software, arrested two men within about a day of a Rs 80 lakh armed robbery in Chandni Chowk and recovered Rs 79.5 lakh","value":{"amount":"Rs 79.5 lakh","what":"stolen cash recovered","who":"the robbed cash courier's firm, R K Enterprises (recovered by Delhi Police from the accused's residence)","estimate":false,"quote":"A raid at [the second man's] residence led to the recovery of Rs 79.5 lakh in cash, a semi-automatic pistol with three bullets, and other incriminating materials.","url":"https://www.rediff.com/news/report/israeli-tech-helps-delhi-police-crack-rs-80-lakh-robbery-2-held/20250319.htm"},"sourceURL":"https://www.rediff.com/news/report/israeli-tech-helps-delhi-police-crack-rs-80-lakh-robbery-2-held/20250319.htm","date_observed_as_stored":"2025-03-19","existing_estimate_flag":false,"existing_comparator_present":false,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":true,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0029","case_id":"AIBD-0207","title":"Restitution to Sheng Siong for stolen wine","summary":"CNA reports that a shoplifter caught after a facial-recognition alert made full restitution of about S$470 to Sheng Siong for the stolen bottles that were not recovered.","monetary_role":"outcome_amount_claim","amount":{"source_text":"about S$470","value":"470","lower":null,"upper":null,"qualifier":"approximately","currency":"SGD","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"recovered_funds","amount_basis":"incremental_change","period":"Thefts between 2 and 11 September 2025. Restitution was made before sentencing on 12 May 2026, on a date not given.","unit":"Restitution from one shoplifter for the bottles not recovered.","beneficiary":"Sheng Siong.","payer":"The shoplifter, a 51-year-old woman.","realization":"reported_recovery","accounting_boundary":"Gross restitution for the bottles not recovered. The three bottles recovered at the interception are excluded, and no costs are deducted.","comparator":"The 19 stolen bottles were worth about S$556. No comparison without the system is given.","ai_attribution":"Staff uploaded her face after finding the thefts on camera footage. The system alerted the manager's phone when she returned, which led to her interception and arrest.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://www.channelnewsasia.com/singapore/shoplifter-caught-sheng-siong-facial-recognition-tech-jail-6115421","title":"Shoplifter identified by Sheng Siong's facial recognition technology gets jail for stealing wine","quote":"made full restitution of about S$470 for the remaining stolen bottles","date":"2026-05-12","primary":true,"written_by":"independent","reviewed_at":"2026-11-01","review_method":"full_text","locator":"Account of the court hearing"}],"limitations":["The date and form of the restitution are not reported.","Staff found the first theft through a stock check and camera footage before the system was used."],"overlap_flags":["The S$556 value of all stolen bottles includes the three bottles recovered in kind. It must not be added to the restitution."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"CNA reports from court that the shoplifter caught after the system's alert made full restitution of about S$470 for unrecovered bottles. The amount is curated as recovered funds, excluding the three bottles recovered in kind.","reviewed_on":"2026-11-01","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"avoided_loss","money_subtype":"restitution for stolen goods","claim_status":"reported_restitution_paid","currency":"SGD","currency_basis":"explicit","unit_basis":"one defendant and series of thefts","time_basis":"Thefts September 2-11, 2025; restitution reported by May 12, 2026.","time_basis_completeness":"explicit_window","monetary_comparator_assessment":"Manual-review versus alert workflow; no monetary counterfactual.","ai_and_value_scope":"AI matched prior thefts and return; human interception, prosecution and restitution followed.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross restitution; investigation and system costs not deducted.","net_benefit_established":false,"comparability_flags":["restitution_transfer","single_series_of_events","different_recovery_dates"],"data_gaps":["Date and proof of restitution","AI contribution relative to human detection","Operating and enforcement costs"],"extraction_note":"The S$470 excludes three recovered bottles; retain distinct cash and in-kind recoveries.","source_review_required":true,"id":"AIBD-0207","title":"Sheng Siong's facial-recognition CCTV, given a shoplifter's face after a 10 September 2025 stock check at its Punggol Central outlet, matches her to six earlier wine thefts and alerts the manager's phone when she returns at noon the next day; staff stop her with three unscanned bottles, police arrest her, she repays the chain and is jailed eight days on 12 May 2026","value":{"amount":"about S$470","what":"restitution paid by the shoplifter for the stolen bottles that were not recovered (three bottles were recovered at the interception)","who":"Sheng Siong","estimate":false,"quote":"The retail manager called the police, and the three bottles of wine were recovered. [The woman] also made full restitution of about S$470 for the remaining stolen bottles.","url":"https://www.channelnewsasia.com/singapore/shoplifter-caught-sheng-siong-facial-recognition-tech-jail-6115421"},"sourceURL":"https://www.channelnewsasia.com/singapore/shoplifter-caught-sheng-siong-facial-recognition-tech-jail-6115421","date_observed_as_stored":"2025-09-11","existing_estimate_flag":false,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":true,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0031","case_id":"AIBD-0184","title":"Estimated Universal Credit overpayments avoided","summary":"The NAO reports DWP’s estimate that its advances model avoided £4.4 million of overpayments over three fiscal years.","monetary_role":"outcome_amount_claim","amount":{"source_text":"Estimated £4.4 million","value":"4400000","lower":null,"upper":null,"qualifier":"approximately","currency":"GBP","currency_basis":"context","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"avoided_loss","amount_basis":"avoided_expenditure","period":"U.K. fiscal years 2022–23 through 2024–25.","unit":"Aggregate Universal Credit advances model estimate.","beneficiary":"U.K. Department for Work and Pensions and public funds.","payer":"DWP benefit expenditure that would otherwise have been overpaid.","realization":"retrospective_estimate","accounting_boundary":"Estimated gross overpayments avoided. DWP could not separate this model’s costs from wider analytics costs.","comparator":"DWP’s estimate of payments that otherwise would have been made; method is not detailed in the cited passage.","ai_attribution":"The model flagged higher-risk applications for staff assessment. The estimate concerns the combined screening and human-review process.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://www.nao.org.uk/wp-content/uploads/2025/10/Tackling-benefit-overpayments-due-to-fraud-and-error.pdf","title":"Tackling benefit overpayments due to fraud and error","quote":"the model generated savings of £4.4 million that would otherwise have been overpaid.","date":"2025-10-22","primary":true,"written_by":"independent","reviewed_at":"2026-10-23","review_method":"full_text","locator":"PDF page 31, paragraph 2.11"}],"limitations":["An independent report relays DWP’s estimate; it does not establish a cost-adjusted AI return."],"overlap_flags":["The model estimate must not be added to broader DWP totals that already include it."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"avoided_loss","money_subtype":"benefit overpayments avoided","claim_status":"retrospective_estimate","currency":"GBP","currency_basis":"symbol_and_UK_context","unit_basis":"DWP Universal Credit advance model","time_basis":"2022-23 through 2024-25","time_basis_completeness":"explicit_window","monetary_comparator_assessment":"Randomized control sample supports risk-identification comparison; the monetary calculation is not set out in the existing fields.","ai_and_value_scope":"Model prioritizes human caseworker review.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross overpayments avoided; account explicitly says model costs cannot be disaggregated.","net_benefit_established":false,"comparability_flags":["gross_costs_not_separable","operator_estimate_in_audit","multi_year_cumulative","distributional_effects"],"data_gaps":["Saving-calculation method","Treatment/control monetary outcomes","Costs if separable","False-positive and delay effects"],"extraction_note":"Independent reporting of an agency estimate should not be mislabeled as an independently verified net saving.","source_review_required":true,"id":"AIBD-0184","title":"DWP's Universal Credit advances machine-learning model saved an estimated £4.4 million from 2022–23 to 2024–25","value":{"amount":"£4.4 million","what":"advances that would otherwise have been overpaid, saved from 2022-23 to 2024-25","who":"Department for Work & Pensions","estimate":true,"quote":"DWP estimates that, from 2022-23 to 2024-25, the model generated savings of £4.4 million that would otherwise have been overpaid.","url":"https://www.nao.org.uk/wp-content/uploads/2025/10/Tackling-benefit-overpayments-due-to-fraud-and-error.pdf"},"sourceURL":"https://www.nao.org.uk/wp-content/uploads/2025/10/Tackling-benefit-overpayments-due-to-fraud-and-error.pdf","date_observed_as_stored":"2025-10-22","existing_estimate_flag":true,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":true,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0032","case_id":"AIBD-0163","title":"Quartz Solar annual imbalance-cost estimate","summary":"Open Climate Fix estimates that Quartz Solar avoids at least £30 million in electricity imbalance costs per year.","monetary_role":"outcome_amount_claim","amount":{"source_text":"At least £30 million per year","value":"30000000","lower":null,"upper":null,"qualifier":"at_least","currency":"GBP","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"cash_saving","amount_basis":"avoided_expenditure","period":"Annualized estimate published November 6, 2025; no completed measurement year is specified.","unit":"Annual electricity imbalance costs.","beneficiary":"Great Britain’s electricity system and its cost bearers.","payer":"Electricity balancing and imbalance settlement participants.","realization":"annualized_estimate","accounting_boundary":"Estimated annual gross operating-cost avoidance, without implementation and ongoing service costs.","comparator":"Previous NESO forecasting tools; the monetary calculation is not provided.","ai_attribution":"Quartz Solar combines machine learning, satellite imagery, and weather data within NESO’s operating workflow. Its developer supplies the monetary estimate.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://www.openclimatefix.org/insights/neso-adopts-ai-solar-forecasting-control-room","title":"National Energy System Operator adopts AI solar forecasting","quote":"helping to avoid at least £30 million in imbalance costs each year","date":"2025-11-06","primary":true,"written_by":"makers","reviewed_at":"2026-10-23","review_method":"full_text","locator":null}],"limitations":["An annualized developer estimate is not a verified cash saving for a completed year."],"overlap_flags":["Future scenarios or cumulative savings cannot be added to this annual rate without a defined period."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The cited monetary statement and its accounting boundary were checked. This is not a certification of net benefit or AI-specific causation.","reviewed_on":"2026-10-23","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"cash_saving","money_subtype":"electricity imbalance costs","claim_status":"annualized_developer_estimate","currency":"GBP","currency_basis":"explicit","unit_basis":"per year at current adoption","time_basis":"Annualized estimate announced November 2025; measured savings interval unspecified.","time_basis_completeness":"annualized_without_window","monetary_comparator_assessment":"Forecast errors reportedly halved; no operator monetary baseline.","ai_and_value_scope":"Maker estimates value of an integrated forecast service; operator confirms deployment, not the saving figure.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross avoided imbalance cost; service and integration costs unstated.","net_benefit_established":false,"comparability_flags":["annualized_not_observed_year","developer_estimate","operational_to_financial_conversion","future_scenario_separate"],"data_gaps":["Calculation from forecast error to cost","Operator confirmation","Observed period","Costs and consumer pass-through"],"extraction_note":"Keep the £30m annual estimate separate from the £150m/year conditional 2035 scenario.","source_review_required":true,"id":"AIBD-0163","title":"Open Climate Fix's Quartz Solar goes live in NESO's control room and may avoid £30m a year in imbalance costs","value":{"amount":"GBP 30 million","what":"imbalance cost avoided per year","who":"Great Britain's electricity system operator NESO and, through balancing costs, consumers","estimate":true,"quote":"Since adoption, Quartz Solar has halved forecasting errors, according to Open Climate Fix, equivalent to saving at least GBP 30 million in imbalance cost per year, with potential to increase to GBP 150 million ($197.0 million) per year in 2035 if government solar capacity targets are met.","url":"https://www.pv-magazine.com/2025/11/07/ai-powered-solar-forecasting-helps-uk-grid-operator-reduce-balancing-costs/"},"sourceURL":"https://www.pv-magazine.com/2025/11/07/ai-powered-solar-forecasting-helps-uk-grid-operator-reduce-balancing-costs/","date_observed_as_stored":"2025-11-06","existing_estimate_flag":true,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"standard","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true},{"id":"DIV-0033","case_id":"AIBD-0181","title":"Revenue recovered in the first results of Ghana's Publican AI","summary":"MyJoyOnline reports that the first results of the Ghana Revenue Authority's Publican AI system showed GH¢15 million recovered from five companies that had falsified trade data.","monetary_role":"outcome_amount_claim","amount":{"source_text":"15 million Cedis","value":"15000000","lower":null,"upper":null,"qualifier":"exact","currency":"GHS","currency_basis":"explicit","price_basis":"Nominal source amount; price year not separately stated."},"additional_amounts":[],"financial_type":"recovered_funds","amount_basis":"incremental_change","period":"Before the launch on 16 January 2026. The recovery dates are not given.","unit":"Recoveries from five companies that had falsified trade data.","beneficiary":"Government of Ghana (Ghana Revenue Authority).","payer":"Five importing companies, not named.","realization":"reported_recovery","accounting_boundary":"Gross amount recovered. The article does not split duties from penalties, and system costs are not deducted.","comparator":"No comparison without the system is given.","ai_attribution":"Publican AI uses machine learning to detect undervaluation and flag high-risk goods for customs officers. The article credits the recovery to the system's initial results.","ai_attribution_status":"ai_component_supported","net_benefit_established":false,"sources":[{"url":"https://www.myjoyonline.com/gra-launches-ai-powered-system-to-boost-revenue-and-curb-customs-fraud/","title":"GRA launches AI-powered system to boost revenue and curb customs fraud","quote":"The system is expected to generate significant revenue for the state, with initial results already showing the recovery of 15 million Cedis from five companies previously falsifying trade data.","date":"2026-01-16","primary":true,"written_by":"independent","reviewed_at":"2026-11-01","review_method":"full_text","locator":"Paragraph after the Commissioner-General's remarks"}],"limitations":["The article does not name the official who gave the figure or publish any record of the recoveries.","Customs officers review the values that the system suggests, and the system's own share of the recovery is not separated."],"overlap_flags":["The revenue gains of 40 to 45 percent from pilots and the later US$300 million in added assessments are separate claims. Neither should be added to this amount."],"curation_status":"curated","source_review_status":"primary_source_checked","case_link_status":"matched","review_note":"The MyJoyOnline launch report says that the system's initial results showed GH¢15 million recovered from five companies that had falsified trade data. It is curated as recovered funds, with the unattributed figure noted as a limitation.","reviewed_on":"2026-11-01","legacy_audit":{"monetary_role":"outcome_amount_claim","money_type":"revenue","money_subtype":"customs fraud recovery","claim_status":"reported_recovery_via_secondary_summary","currency":"GHS","currency_basis":"explicit","unit_basis":"program recovery to date","time_basis":"Reported January 2026; recovery interval unspecified.","time_basis_completeness":"incomplete_window","monetary_comparator_assessment":"Assessment-time comparison only; claimed revenue-gain percentages refer to pilots.","ai_and_value_scope":"AI analytics claim and recovery appear in a registry summary; source account does not establish mechanism or incremental share.","gross_net_status":"not_explicitly_established","gross_net_interpretation":"Gross recovery claim; costs and refund/reversal adjustments are absent.","net_benefit_established":false,"comparability_flags":["source_chain_needs_review","recovery_period_missing","pilot_vs_rollout_timing"],"data_gaps":["Original GhanaWeb source and authority confirmation","Recovery dates and accounting definition","AI role","Costs"],"extraction_note":"The 40-45% pilot revenue claim and the GH¢15m recovered need separate claim and cohort identifiers.","source_review_required":true,"id":"AIBD-0181","title":"Ghana Revenue Authority launches Publican AI trade analytics, which may have recovered GH¢15 million","value":{"amount":"GH¢15 million","what":"recovered from fraudulent importers","who":"Government of Ghana (Ghana Revenue Authority)","estimate":false,"quote":"The system has already enabled the government to recover GH¢15 million from fraudulent importers and aims to address significant revenue leakages in Ghana's customs operations.","url":"https://www.myjoyonline.com/gra-launches-ai-powered-system-to-boost-revenue-and-curb-customs-fraud/"},"sourceURL":"https://www.myjoyonline.com/gra-launches-ai-powered-system-to-boost-revenue-and-curb-customs-fraud/","date_observed_as_stored":"2026-01-16","existing_estimate_flag":false,"existing_comparator_present":true,"source_verification_status":"not_reopened_or_verified_in_this_audit","reviewed_existing_fields":["value","title","metric","comparator","account","ai_role","date_observed"],"source_review_priority":"high","unit_basis_present":true,"amount_specific_measurement_window_fully_explicit":false,"original_quote_and_source_link_present":true},"eligible":true}]