Key Points
- Deloitte, KPMG, EY and PwC have all faced allegations that generative artificial intelligence (AI) was used extensively in reports presented as professional research or thought leadership.
- The allegations have raised questions about the reliability of consulting research, the accuracy of published references and the quality-control procedures used by major professional-services firms.
- Deloitte agreed to repay AU$440,000 to Australia’s Department of Employment and Workplace Relations after a government report was corrected to remove more than a dozen references and revise its text.
- AI-detection company GPTZero reported that 40 of 45 sources in a KPMG report published in October 2025 were hallucinated, while half of the factual claims it examined were allegedly false or misattributed.
- In May 2026, GPTZero alleged that a cybersecurity report published by EY Canada was 72% AI-generated and identified problems with 16 of its 27 citations.
- GPTZero subsequently examined four PwC studies published between 2024 and 2026 for clients in the Middle East. Its software estimated an 84% chance that the report Transforming Governance consisted entirely of generative text.
- The PwC report promoted a framework called Citizen Pulse and made claims about government use in Denmark, Saudi Arabia, the United States and Australia. GPTZero said it found no public evidence supporting those claims.
- PwC Middle East said it was updating a limited number of supporting citations, while KPMG International said its report had been removed and the circumstances surrounding publication were being reviewed.
- Matt Parry, chief executive of technology consultancy The Future Collective, warned that inaccurate corporate reports could be recycled by AI systems and subsequently repeated as though their claims had been independently verified.
- The controversy raises broader questions about professional accountability, the use of generative AI in consulting and the credibility of research used to attract potential clients.
Deloitte, KPMG, EY and PwC, the four largest global professional-services networks commonly known as the Big Four, have faced allegations that generative AI was used to produce substantial portions of research reports published as thought leadership. In a report published on 12 August 2026, Consultancy.uk described a series of incidents involving inaccurate references, questionable factual claims and concerns about the verification of AI-generated material. Investigations by AI-detection company GPTZero formed part of the allegations concerning KPMG, EY and PwC, while Deloitte previously agreed to repay AU$440,000 following corrections to a report commissioned by the Australian government. The cases have intensified scrutiny of the firms’ research practices, with potential implications for client confidence, professional accountability and the reliability of information subsequently used by other AI systems. The allegations differ in scope, and AI-detection estimates alone do not establish that every disputed passage was generated by AI or that the firms deliberately misrepresented their work.
Why Are the Big Four Facing Scrutiny Over AI-Generated Research?
The Big Four—Deloitte, KPMG, EY and PwC—use reports, surveys and industry analysis to demonstrate expertise in particular sectors and encourage prospective clients to engage their consulting services.
These publications can cover topics such as digital transformation, cybersecurity, corporate governance, customer experience and emerging technologies. They are often designed to provide practical recommendations and demonstrate a consultancy’s understanding of business challenges.
According to Consultancy.uk, the growing use of generative AI has created incentives for firms to automate parts of this work, potentially reducing the time and cost required to produce reports. However, the publication argued that the incidents involving the four firms had raised questions about whether the pursuit of efficiency and AI-related credentials had come at the expense of research quality.
The central concern is not simply whether AI tools were used. Generative AI can support research, drafting and data analysis, but its outputs can contain fabricated references, incorrect facts and misleading attributions.
For consulting firms, those problems become particularly significant when research is published under a recognised corporate name and presented as evidence of professional expertise.
The allegations also raise questions about the distinction between using AI as an assistance tool and relying on generated material without sufficient verification by qualified researchers.
What Happened in Deloitte’s Australian Government Report?
Deloitte was the first of the four firms discussed in the Consultancy.uk report to face a major public controversy involving alleged AI-generated inaccuracies in a government report.
The incident concerned a report commissioned by Australia’s Department of Employment and Workplace Relations. According to Consultancy.uk, the report contained fabricated citations, incorrect footnotes and a quotation attributed to a Federal Court judgment that did not exist as presented.
Following the controversy, Deloitte agreed to repay AU440,000,equivalenttoapproximatelyUS291,245 at the time cited by the publication.
The department subsequently uploaded a corrected version of the report. Consultancy.uk reported that the revision removed more than a dozen references and rewrote portions of the text.
Australian officials nevertheless maintained that the substance of the report remained intact.
The distinction is important: the corrections and repayment established that changes were made to the commissioned work, but the department’s position did not amount to a finding that the report’s entire analysis was invalid.
The incident nevertheless demonstrated the risks of publishing research containing references that readers cannot reliably verify. Government departments and other clients depend on consultants to provide evidence-based analysis, particularly when reports may inform policy, operational decisions or public expenditure.
For consulting businesses, the episode also highlighted the importance of checking references and quotations before delivering reports to clients.
What Did GPTZero Find in KPMG’s Thought Leadership Report?
KPMG faced separate allegations after GPTZero examined its October 2025 report, Total Experience: Redefining Excellence in the Age of Agentic AI.
As reported by Consultancy.uk, GPTZero’s forensic review found that 40 of the report’s 45 sources had allegedly been hallucinated by generative tools. The investigation also claimed that approximately half of the factual statements examined were false or attributed incorrectly.
One example involved a reference described as a 2019 East Japan press release that was cited as evidence of agentic AI use. The report noted that agentic AI only entered public discourse in 2024, raising questions about the suitability of the reference for supporting the claim.
The investigation also identified problems with the attribution of authorship. According to Consultancy.uk, the language model repeatedly confused the subjects of articles with the people who had written them, potentially making sources appear to provide support that they did not actually offer.
These findings concern the report examined by GPTZero and should not be interpreted as evidence that all KPMG research contains similar problems.
KPMG International responded by saying that it took the accuracy and integrity of its published content seriously. The firm said the report had been removed and that it was reviewing the circumstances surrounding its publication.
The response indicated that the publication was being reviewed, but did not, by itself, establish the precise extent of AI involvement or resolve every allegation concerning the report.
Why Was EY Canada’s Cybersecurity Research Challenged?
EY became the subject of further scrutiny in May 2026 when GPTZero examined a report published by EY Canada.
The study, titled Points of Attack: Uncovering Cyber Threats and Fraud in Loyalty Systems, focused on cybersecurity threats and fraud involving loyalty programmes.
According to Consultancy.uk, GPTZero alleged that 72% of the report was AI-generated. Its investigation also identified problems with 16 of the 27 citations.
The publication reported that the disputed references included links or purported sources associated with organisations and publications such as Forbes, McKinsey & Company, Gartner, TechCrunch and WIRED. Some citations were described as broken URLs, while others allegedly referred to sources that could not be found.
The distinction between a broken link and a source that never existed is relevant. A link may stop working for reasons unrelated to whether a source originally existed, while a fabricated citation presents a more fundamental problem for the credibility of research.
The reported findings therefore raised questions about whether the evidence cited in the cybersecurity report could be independently checked by readers.
However, the AI-generation percentage attributed to GPTZero is an estimate from a detection tool, not conclusive proof of authorship. Establishing how individual passages were produced would require a broader examination of the document’s development, source material and editorial process.
The episode nevertheless added EY to the group of firms facing scrutiny over the reliability of published research.
What Did GPTZero Allege About PwC’s Citizen Pulse Framework?
PwC became the fourth firm discussed in the Consultancy.uk report after GPTZero examined four studies published between 2024 and 2026 for clients in the Middle East.
The investigation focused particularly on Transforming Governance, a report that promoted a PwC framework called Citizen Pulse.
According to Consultancy.uk, GPTZero’s detection software estimated an 84% probability that the report consisted entirely of generative text. The publication did not present that figure as definitive proof that every part of the report had been written by AI.
The more consequential allegation concerned claims about the framework’s use. The report reportedly stated that governments in Denmark, Saudi Arabia, the United States and Australia were using Citizen Pulse.
GPTZero said it could find no public evidence confirming that the framework was being used by those governments or that the corresponding arrangements had taken place.
The investigation therefore raised two distinct questions: how the report was produced and whether the claims about the framework’s adoption were supported by evidence.
Consultancy.uk also quoted Om Ogale of GPTZero, who questioned whether Citizen Pulse was being described as an established PwC product or a more general conceptual framework. Ogale argued that the report moved between discussing potential benefits and making claims about existing use, which he described as a symptom of poorly prompted large language model output.
The distinction matters because a framework proposed in a consulting report is not necessarily the same as a product that has been implemented by a government. Claims of adoption should be supported by evidence that readers can independently examine.
How Did PwC and KPMG Respond to the Allegations?
PwC Middle East did not issue a direct denial of using generated content in the report discussed by Consultancy.uk.
Instead, the firm said it took the accuracy of its published research seriously and was updating a limited number of supporting citations in the reports mentioned.
KPMG International issued a separate statement following the allegations concerning its own report. It said the report had been removed and that the circumstances surrounding its publication were under review.
Both responses referred to the accuracy or integrity of published material and the need to review the circumstances surrounding the reports.
However, the statements differed in their specific actions: PwC described updates to supporting citations, while KPMG said its report had been removed.
The statements do not establish that the firms admitted every allegation made by GPTZero. Nor do they independently establish that the reported AI-generation estimates were accurate.
The responses do, however, illustrate the importance of reviewing research when questions emerge about its sources and reliability.
For clients, the key issue is whether published work can withstand independent scrutiny and whether the consultancy responsible has adequate procedures for correcting mistakes.
How Could AI-Generated Errors Damage Trust in Consulting Firms?
Thought leadership occupies a distinctive position in the consulting industry. Although such reports may not directly generate revenue, they can help firms demonstrate expertise, attract prospective clients and establish credibility in specialist markets.
The reports may also influence how businesses understand emerging technologies, assess risks and formulate strategies.
When a publication contains fabricated citations or unsupported claims, readers may question the quality of the research as well as the review processes that allowed it to be published.
This creates a reputational risk that extends beyond an individual document. Clients may ask whether similar verification standards are applied to the work they commission and whether senior specialists review important findings before they are delivered.
The incidents do not establish that the Big Four routinely provide unreliable client services. However, they raise questions about whether the quality-control arrangements used for public-facing research are sufficient when generative AI is involved.
Organisations using AI in research and consulting need to distinguish between drafting assistance and evidence verification. A generated reference may look credible without corresponding to a real publication, while a fluent explanation can conceal an incorrect factual claim.
Effective review therefore requires more than checking whether a document reads well. Researchers and editors must verify sources, confirm quotations, assess the evidence supporting claims and ensure that the conclusions follow from the available information.
For firms whose reputations depend on professional judgement, these checks remain central to maintaining client confidence.
Why Could AI-Generated Consulting Reports Spread Inaccurate Information Further?
The controversy also raises concerns about how inaccurate information can spread beyond the original report.
AI systems may retrieve information from publicly accessible corporate reports, research papers and media publications. If an inaccurate claim appears on a trusted website or in a widely circulated document, another system may encounter and reproduce it when generating a later answer.
That process can make an unsupported claim appear more credible, particularly if it is repeated across several sources without proper verification.
Consultancy.uk highlighted this issue through comments from Matt Parry, chief executive of technology consultancy The Future Collective.
Parry warned that AI systems could retrieve, cite and repeat information from these reports, allowing claims about products or government adoption to circulate as though they had been corroborated.
His concern was that repeated references to the same original claim could create an illusion of independent confirmation. In reality, multiple AI-generated answers might rely on the same underlying material.
This is a potential weakness in information ecosystems that rely on retrieval and summarisation. Repetition does not necessarily establish accuracy, and the appearance of several supporting references can be misleading if those references ultimately trace back to one flawed source.
The issue is especially relevant to corporate strategy, where executives increasingly use AI tools to summarise market developments, investigate competitors and explore investment opportunities.
Businesses need to consider the provenance of information, verify material claims against original evidence and avoid treating the presence of a citation as proof that a statement is correct.
What Does the Controversy Mean for Corporate AI Governance?
The cases involving Deloitte, KPMG, EY and PwC demonstrate why organisations need clear rules governing the use of generative AI in professional work.
AI can support research and improve efficiency, but businesses must establish who is responsible for checking its output and what evidence is required before generated material is published.
For consulting firms, appropriate controls may include mandatory source verification, human review of factual claims, checks on quotations, documentation of AI use where relevant and clear procedures for correcting published errors.
The appropriate safeguards will depend on the type of work being produced. A preliminary brainstorming document may carry different risks from a report commissioned by a government department or used to support a major corporate decision.
Organisations should also ensure that responsibility for the final product remains clear. Using an AI tool does not remove the need for professional judgement or accountability for the work delivered to a client.
These issues are relevant to Artificial Intelligence (AI), Management, and Media and Public Relations, where professionals must assess how automated content is developed, how information is verified and how organisations communicate when errors occur. They also connect to Legal, Contracts and Procurement, particularly where commissioned reports, contractual obligations and the accuracy of deliverables are involved.
What Happens Next for the Big Four’s Research and Publishing Practices?
The longer-term implications will depend on how the firms respond to the allegations, whether further corrections are made and whether their quality-control processes change.
The incidents discussed by Consultancy.uk involve different reports, different clients and different types of alleged error. They should not be treated as identical cases or as proof that all AI-assisted research is unreliable.
Nevertheless, the repeated appearance of concerns involving all four major consulting networks has placed greater attention on the standards used to produce and verify thought leadership.
For clients, the practical question is whether consulting research can be independently checked and whether the evidence supports the recommendations being made.
For the firms themselves, maintaining credibility will require more than demonstrating that they use advanced technology. Their publications must also meet expectations for accuracy, transparent sourcing and professional review.
The debate is therefore broader than whether AI should be used in consulting. It concerns how businesses can use the technology without undermining the reliability of the work they publish or the confidence that clients place in their expertise.
The reports examined by GPTZero have raised significant questions, but the precise extent of AI involvement and the status of individual disputed claims must be assessed on a case-by-case basis. The central lesson is that efficiency gains from automation do not replace the need for evidence, editorial scrutiny and accountability.
As companies expand their use of generative AI, the ability to verify information and maintain reliable governance will remain an important part of responsible business practice.