Are We Running on AI Vibes These Days?

Note, this is still a new publication and it’ll take me time to get my bearings right. I’d appreciate your feedback, comment below or send me a message! In the meantime, enjoy this long-form article about AI causing issues and consequences.

A recent research article from the investigators at GPTZero revealed that a report published in October 2025 by KPMG about agentic AI contained many errors and AI hallucinations and “vibe citations” that it was pulled down from their website. Last month, EY pulled a report due to AI hallucination issues.

For reference, agentic AI is a form of artificial intelligence that uses generative AI tools (like ChatGPT, Claude, Gemini, or CoPilot) and makes decisions and actions similar to a human. In addition, AI hallucinations are outputs made by AI tools that are non-sense or inaccurate, sometimes confidently passing the information as accurate. Lastly, a vibe citation is a citation that contains hallucination errors created by the generative AI tool.

The researchers from GPTZero investigated KPMG’s Total Experience: Redefining Excellence in the Age of Agentic AI report and identified that 5 out of 45 citations were fully accurate. They suspect no human checked the citations and mentioned that “the poor quality of the footnoted citations” was enough that they “didn’t attempt an exhaustive verification of every claim.”

Notable findings identified by GPTZero about the report include:

• Fabricated claims about companies using technology that didn’t exist yet, even fabricated claims a chatbot from an airline could change flight bookings.

• Invented capabilities of a company, confusing two separate companies’ AI capabilities.

• Misinformation is spreading because LLMs are citing the report and articles (tied to the report) with inaccurate information.

GPTZero researchers say that vibes have consequences. They say that statistics and claims from that report “have been repeated and recycled.” Another issue is that other AI tools can cite the inaccurate content further creating a spiraling misinformation loop because those articles and posts continue to present incorrect information from the report. It can erode the quality of accurate information because it can be harder to trace where the inaccuracy gets introduced. If this happens more often, will our information become inaccurately unusable thanks to vibe citations?

A related and similar issue comes from programmers who are “vibe coding” where security issues, malware, and programming flaws could be introduced into an application or service that millions of people rely on. Like vibe citations, vibe coding inaccuracies, inefficiencies, or errors may not show up at first until someone digs into the code and finds the issues and corrects them.

While it is astounding to hear of a major consulting agency making such a massive error relying too much on AI, this can happen to anyone, but it shouldn’t happen so often. We saw this years ago where a judge tossed out a case because ChatGPT cited non-existent cases. Did we learn?

Here’s my lessons that you can take away from here:

• Understand that AI-generated research or content can contain inaccurate content presented as accurate

• If you use AI to fetch information, make the effort to verify the claims from the sources

• Do the research yourself, take the time to collect and verify the information, bring together your experience and research skills

• You are 100% responsible for any content you create, even if some of the work is aided by AI tools

I’ll go further with my final thoughts here: you own your own destiny with artificial intelligence.

You can use AI to do the work for you and not learn a single thing OR you can use AI in strategic ways that enable you to use your experience and research skills, combined with your own rich human curiosity.

Like what you read? I ask that you encourage a friend or colleague to subscribe today.

Share Cyber Communicator | Roger Renteria

Help me identify cybersecurity topics that you want to hear from me! Consider taking a moment to take my survey to balance my coverage of topics that matter to you.

Take the Topics Survey

Note: headline image was created by generative AI using the following prompt: “create an image of an AI hallucination that contains text, business report, and make it very much like it came from the depths of the universe”