Deloitte Corrects AI-Generated Report on Targeted Compliance Framework
Deloitte has issued a revised report on the Targeted Compliance Framework, correcting earlier inaccuracies. The initial report, generated using Microsoft's GPT-40 language model, contained false information and hallucinated data.
Deloitte's use of Microsoft's GPT-40 for the review report led to significant errors. The system misinterpreted laws and guidelines, resulting in wrongful penalties for at least 1371 Australians over five years. Deloitte has since refunded part of the fee for the report due to these AI-generated inventions.
The Targeted Compliance Framework, which automatically penalizes social benefit recipients based on suspected non-compliance, lacks transparency in its appeal procedures. There's no documentation of program logic or performance standards, making it difficult to trace and validate decisions. Moreover, the system lacks traceability, validation, risk management, and oversight in its processes.
The revised Deloitte report addresses the earlier issues, with no evidence supporting the claim that academics from Lund University or the University of Sydney were cited as authors of non-existent studies. The use of AI in such critical systems requires robust validation and oversight to prevent similar incidents in the future.
Read also:
- C3IC Publishes 'This is the Taiwan We Love' Guide to Empower Citizens
- Reporter of Silenced Torment or Individual Recording Suppressed Agony
- Solar panel troubles on rooftops
- EPA Administrator Zeldin travels to Iowa, reveals fresh EPA DEF guidelines, attends State Fair, commemorates One Big Beautiful Bill
