NetGoodIndexSubmit a correction

Harm Ledger · retracted · Computer Science

Deloitte refunds part of an Australian government report after GPT-4o hallucinations

Deloitte Australia’s July 2025 welfare-compliance review for DEWR, produced with Azure OpenAI GPT-4o, contained fabricated citations and a misquoted judgment; a corrected version was issued and Deloitte agreed to repay the final contract installment.

4 Jul 2025Tier 2 Significant HarmMethodology 0.1

Current score

−0.09

3 base · Significant Harm (tier 2 of 5, 3 pts)
× 0.3500 attribution · Material acceleration
× 0.8500 evidence · Multiple independent confirmations
× 0.8000 realization · Independently validated or deployed
× 0.3000 durability
Event-level product before credit split: 0.21

Corrupt official evidence in a paid public report is notable institutional harm (tier 2). Human-led consulting with AI assistance (0.35). Independent academic catch plus government correction (0.85). Realization is a published then withdrawn/corrected report. Residual harm falls after correction.

What happened

The original report listed references that do not exist and misspelled Justice Jennifer Davies as Davis. After academic and press scrutiny, DEWR posted a revised document disclosing the model. Deloitte and the department said recommendations were unchanged. The event is a serious correction of an official AI-assisted work product, treated here as retracted for scoring of the original report’s reliability.

Model attribution

−0.09

GPT

Azure-hosted model used in Deloitte’s technical workstream that generated fabricated references.

Deloitte disclosed GPT-4o use; humans delivered the report without adequate verification.

Attribution 0.3500 · Credit share 40% · OpenAI

Claims

  • The original DEWR report contained fabricated academic references and a misquoted judicial passage later removed in a corrected version.

    outcome · supported

  • Deloitte agreed to repay the final installment of the roughly A$440,000 contract.

    outcome · supported

Sources

independent sources

Revision history

  • 13 Sep 2026 · 0.00 → 0.09

    Initial adjudicated seed score under methodology 0.1.