EY Canada published a cybersecurity report and most citations were hallucinated
A recent investigation found that a high proportion of citations in an EY Canada cybersecurity report were fabricated by generative AI, raising questions about the firm’s quality controls and the broader use of AI in professional research and consulting. Commenters argue that many organizations are delegating substantive work to AI without adequate expert vetting, often because those experts are overstretched or sidelined, which can make AI a net negative when accuracy matters. The episode is framed as symptomatic of deeper structural problems: consulting reports that exist mainly for executive “cover,” corporate incentives to optimize for cost over rigor, and a growing reliance on “vibe” outputs that few people read closely but which still shape decisions and training data.
Site usability and presentation
- Strong consensus that the investigative site is almost unreadable due to extreme scroll hijacking, parallax effects, forced pauses, and dual scrollbars.
- Both mobile and desktop users report stuttering, stalls, getting thrown back to the top, and being unable to reach key sections.
- Many say they abandoned the article purely because of the UX, calling it “user-hostile” and wondering why such heavy JavaScript was used.
- Several ask for a plain-text version or rely on browser reader modes, which only partially work and sometimes hide images containing sources.
EY report and hallucinated citations
- Readers note the headline stat about hallucinated citations rises as you scroll; some initially saw ~26%, but the final figure reported is ~60%.
- Many view this as an embarrassment for a premium consulting/audit brand whose business is supposed to be trusted analysis.
- Some argue it confirms that such reports were always mostly “box-ticking” or CYA artifacts rather than serious research.
- There’s concern that chatbots and aggregators will ingest these flawed reports and propagate them as fact.
AI usage, vetting, and hallucinations
- Broad agreement that AI outputs are often not vetted by domain experts, due to overload, cost pressure, or laziness.
- Multiple professionals (especially in law and software) say reviewing AI output can take longer than doing work from scratch, especially where precision and citations matter.
- Others describe productive patterns: using AI for research, gap analysis, or code review rather than primary drafting.
- Discussion of multi-LLM “council” approaches to catch hallucinations: seen as helpful but slower, more expensive, and still not a substitute for human oversight.
Consulting industry and Big Four critiques
- Many see the EY incident as evidence that large consulting firms provide generic, low-value work and are now “vibe-citing” with LLMs.
- Several argue these firms mainly launder blame and justify pre-decided strategies for executives and boards.
- Some describe long-standing issues: overworked or inexperienced staff, declining audit quality, and advice that’s risk-averse and non-actionable.
Work, incentives, and broader concerns
- Commenters highlight systemic incentives: “generate now, review later” mirrors tech-debt patterns and short-termism.
- There’s anxiety about AI hollowing out skilled roles, turning experts into overworked proofreaders of “AI slop.”
- Others think AI will expose weak management and could eventually diminish the need for traditional consultants.