About a month ago, observers first raised concerns that a report Deloitte had produced for the Australian federal government included material that could not be traced to real sources. Multiple instances of fictitious citations and misattributed quotes were flagged: references to papers and judicial rulings that simply do not exist, and a quote supposedly from Justice Davies that appears nowhere in the relevant court judgment. That suspicion spurred further scrutiny, and Deloitte just admitted it had used generative AI in producing the report, prompting it to partially refund the government and issue corrections.
That incident is not isolated. It exemplifies a deeper and growing problem in how AI is being used in professional contexts: the normalization of “AI slop”, that is, machine-generated content that is polished on the surface but flawed underneath, and which slips through because it appears credible. The Deloitte case should serve as a warning: capability without oversight is a liability.
In the Deloitte report’s original version, academics noted that several of the cited works by scholars at the University of Sydney and other institutions do not exist, and that the legal quotes were distorted or fabricated. Deloitte defended its findings but acknowledged the need to correct references. When the revised version was posted, it included disclosure of AI assistance via Azure OpenAI GPT-4o. The firm maintained the substance of the report but conceded that footnotes and citations had errors.
Meanwhile, in a different neighborhood of AI use, OpenAI’s Sora 2 model has reignited debates about how generative media enters public discourse. The tool allows users to generate short videos from prompts, potentially including uploaded faces and voices. Some of its early outputs were mistaken by viewers for real footage, prompting criticism over labeling, intellectual property, and the boundary between synthetic and authentic content. Vinod Khosla, an early investor in OpenAI, dismissed the critics of Sora 2 as “tunnel vision creatives” and urged that viewers themselves should assess the output.
The common thread between Deloitte’s misstep and the Sora 2 controversy is not technical failure; it is institutional failure to govern AI-generated work. In Deloitte’s case, AI-generated claims made it into a client deliverable without sufficient verification. In Sora 2’s case, synthetic media is already spilling into public feeds without clear attribution or accountability. In both cases, the risk lies in the slop being indistinguishable from plausible, polished output, until someone digs.
That slop is not just aesthetic or academic. Recent research on workplace AI use quantifies its cost. The Harvard Business Review (in collaboration with BetterUp Labs) found that about 40 percent of surveyed employees had encountered AI-generated content that required editing or correction. On average, those instances consumed nearly two hours of additional work, and recipients reported lower trust in both the content and the person who deployed it. The researchers referred to this as an “invisible productivity tax.” That phenomenon can be thought of as “workslop”, generative output that appears finished but requires extra labor to salvage.
The Deloitte incident is a scaled-up version of workslop entering a government contract. In both cases, what looks polished can conceal gaps, distortions, or outright hallucinations. As generative models become more fluent and persuasive, the burden shifts increasingly to humans to catch what the machines get wrong.
That burden is not evenly distributed. In the Deloitte case, Labor Senator Deborah O’Neill called the episode a “human intelligence problem,” urging that “anyone looking to contract these firms should be asking exactly who is doing the work they are paying for, and having that expertise and no AI use verified.” The Guardian also reported that the revised report added a methodological note disclosing the use of a generative AI tool chain licensed through Azure OpenAI GPT-4o.
If more organizations adopt generative models without updating their verification pipelines, the volume of AI slop could outpace their capacity to catch it. As AI governance scholars note, many organizations still lack structures like audit trails, stakeholder accountability, or internal review boards. Only a fraction of enterprises have enterprise-wide oversight mechanisms; one survey suggests that just 18 percent of organizations have a centralized council authorized to make decisions on responsible AI governance.
The core issue is not that AI is unusable, but that AI use has outpaced quality control. In practice, many organizations emphasize granting access to generative tools (encouraging experimentation) rather than instituting rigorous protocols for verification or disclosure. Those tools are increasingly being deployed in client-facing or public contexts without a parallel discipline of review. Because generative models produce fluent, polished output, it is easy for errors or fabrications to masquerade as credibility.To mitigate that risk, output intended for client delivery, public messaging, or policy work should be audited for factual accuracy, contextual consistency, and source integrity.
Governance literature recommends logging where AI is used, requiring human sign-off, and building review boards or committees to examine AI-generated content before release. In practice, some organizations are exploring algorithm review boards or integrating AI oversight into existing compliance functions. Clearly, this hasn’t happened fast enough. The lesson from these recent cases is clear: the technology may be advancing fast, but the systems around it are lagging.
