This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Academics submitted a report to an Australian parliamentary inquiry containing false allegations against major consultancy firms, including KPMG and Deloitte. The false information was generated by Google Bard. The academics issued an apology after the firms identified the inaccuracies, highlighting the risks of using AI to generate content for formal or high-stakes proceedings.
Australian academics reportedly used Google Bard AI to generate case studies for a parliamentary inquiry, leading to false allegations against major consultancy firms. The AI-generated misinformation prompted an apology from the academics, causing reputational harm for all parties involved and raising concerns about the reliability of AI tools in producing accurate and unbiased information.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that inadvertently generate or spread incorrect or deceptive information, which can lead to inaccurate beliefs in users and undermine their autonomy. Humans that make decisions based on false beliefs can experience physical, emotional or material harms
Also classified as
The academics overrelied on the AI tool, assuming its authoritative-sounding output was accurate without performing basic fact-checking.
The AI system generated false and misleading allegations of financial misconduct, which were then published in a public parliamentary submission.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The reputational risk was caused by Google Bard generating false allegations, which was an unexpected and unintentional outcome of the academics using the tool to draft their submission.
Added categories peak at 2: Minor (Epistemic Harm, Direct)
Australian academics unintentionally submitted fabricated case studies generated by Google Bard to a parliamentary inquiry, falsely accusing KPMG and Deloitte of misconduct. The incident highlights minor risks to government functions through the potential pollution of official public records, though it was quickly corrected with no lasting national security impact.
The academics used Google Bard as an assistant to generate case studies and text for their parliamentary submission, which required human oversight and verification that was not properly executed.
Limited Risk: Google Bard is a general-purpose chatbot and content generation tool, which falls under the limited risk category requiring transparency about its AI nature.
The AI system was used by the academics to generate written case studies and text for their formal parliamentary submission.
The report provides clear, direct, and uncontradicted accounts of the incident, including statements from the affected firms, the Senate committee chair, and the academic who took responsibility.
None
Technology
AI Hallucination
Google Bard generated realistic but false case studies of misconduct.
Authoritative-Sounding Output
The LLM generated highly confident text that masked its factual inaccuracy.
Data Inputs
Lack of Source Verification
The model lacked real-time verification of historical audit records.
Human Factors
Overreliance on AI Output
Academics trusted the AI-generated facts without secondary verification.
Lack of AI Literacy
The lead academic had only started using Google Bard that same week.
Process and Methods
Inadequate Fact-Checking
The authors failed to verify the generated claims before submission.
Flawed Submission Drafting
The research process integrated unverified AI outputs directly into draft.
Management
Lack of Co-Author Oversight
Co-authors did not review or verify the lead author's AI-generated content.
Regulatory Environment
Abuse of Parliamentary Privilege
Privilege protected false accusations from immediate legal defamation checks.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
