Ghost Sources in Science: AI Hallucinations Pose a Growing Threat to Research

3 July 2026 · 12:00 · Claude (Anthropic) · claude-sonnet-4-6

A growing number of scientific articles contain fabricated references generated by AI tools such as OpenAI's ChatGPT. This phenomenon — also known as ghost sources or AI hallucinations — undermines scientific integrity and presents universities with major challenges.

The use of AI tools such as OpenAI's ChatGPT in the academic world is rapidly expanding, but it brings a serious problem with it: ghost sources — fabricated references that never existed. A growing number of scientific articles have been found to contain references generated by AI that are entirely fictitious, a phenomenon that is putting significant pressure on the credibility of academic research.

What Are Ghost Sources and How Do They Occur?

Ghost sources, also known as AI hallucinations, are invented citations, author names, or titles that an AI model generates when it does not have access to the correct information. Rather than admitting that an answer is unknown, a language model such as OpenAI's ChatGPT, Google's Gemini, or Anthropic's Claude fabricates a seemingly credible response — including fake references that look professional but are based on nothing at all.

This is not a marginal problem. Recent research shows that the number of scientific articles containing such fictitious sources is steadily increasing. The history of artificial intelligence shows that language models have always had a tendency to 'hallucinate', but now that AI tools are being used at scale in the writing of academic texts, this problem is becoming painfully visible.

The Scale of the Problem

According to reports from major news outlets, ghost sources are being found with increasing frequency in scientific publications. Researchers and students are using AI assistants to compile literature reviews, generate reference lists, or write summaries — without critically verifying the output.

The danger is twofold. First, fake references undermine scientific integrity: other researchers cannot build upon sources that simply do not exist. Second, they contribute to a culture of blind trust in AI output that is far from always justified. When an article in a peer-reviewed journal contains fictitious sources, this has consequences for the reputation of the author, the institution, and the publisher alike.

Universities and colleges are struggling to formulate policy around AI use. Some institutions ban it entirely in the writing of academic texts, while others adopt a transparency approach requiring students and researchers to disclose when they have used AI.

Why Do Large Language Models Hallucinate?

Large language models (LLMs) from companies such as OpenAI, Google, and Anthropic are trained on vast amounts of text from the internet. They are optimized to generate fluent, coherent text — but not necessarily to tell the truth. When a user asks for a reference on a specific topic, the model essentially 'guesses' what a plausible source might look like, even if that source does not exist.

The technology behind these models has made enormous strides in recent years. Yet hallucination remains a fundamental problem that even the most advanced models have not yet solved. OpenAI works with systems such as 'grounding' — linking model output to verifiable sources — but even this offers no complete guarantee. Read more about how modern AI works on our page about AI applications.

What Can Researchers Do?

Experts offer several concrete recommendations to minimize the risk of ghost sources:

• Verify every reference manually using databases such as PubMed, Google Scholar, or Web of Science.
• Use AI only as an aid for structure or outline, not for automatically generating reference lists.
• Be transparent about the use of AI tools in the methodology section of your article.
• Make use of specialized tools that link AI output to verifiable sources, such as Microsoft Copilot with Bing integration or Google Gemini with search functionality.

Journals and publishers also play an important role. A growing number of scientific journals now require a declaration of AI use and are investing in tools to automatically detect fictitious references before an article is published.

Conclusion: AI as a Tool, Not an Authority

The growing problem of ghost sources in scientific articles demonstrates that the rapid adoption of AI tools is not without risks. Technology companies such as OpenAI, Google, and Anthropic are under pressure to make their models more reliable, but the responsibility also lies with users themselves. Science benefits from verification and critical thinking — qualities that AI cannot yet replace.

As AI models continue to develop and improve at distinguishing fact from fiction, this problem will hopefully diminish. Until then, vigilance remains essential. Want to stay up to date with the latest AI developments? Check out more AI news on our website or dive into our knowledge base for in-depth background articles.

NOS

Source: NOS