AI hallucinations have seen plague in science: false references come into definite knowledge

Maxim Topaz, a professor at Columbia University, found in his own article a reference he had never written: a completely manufactured source, inserted by an artificial intelligence tool used by coauthors. This is not an isolated case. According to the report of Yahoo News, the proliferation of AI hallucinations — false references that seem legitimate — 'vertically' in 2024, infiltrating the work of experts and gaining consolidated knowledge status.

How do AI hallucinations fool researchers?

Hallucinations occur when language models generate information that seems plausible, but are completely invented. In the academic case, this translates into citations of articles that never existed, fictional authors and invented data. Topaz reported that the phenomenon affects not only literature reviews, but also articles submitted to journals, where false references go unnoticed by the evaluators. The impact is cumulative: each false reference accepted becomes part of the scientific record, contaminating future research that cites it. This is especially dangerous in areas such as medicine and public policies, where evidence-based decisions can be compromised.

Impact of AI hallucinations on Brazilian science

Brazil, with its expanding graduate system and pressure for publications, is particularly vulnerable. Brazilian researchers, often overloaded, can use AI tools to accelerate bibliographic reviews — without manually checking each reference. The growing adoption of AI in the country, as pointed out by a recent alert on cyber attacks with AI, shows the rapid pace of technology, but the human curatorship did not follow. If a Brazilian article accepts a false reference, it can spread through future citations, inflating knowledge that does not exist. Institutions such as CAPES and CNPq need to include source verification guidelines in their evaluation criteria.

Strategies to contain false references in articles

The solution is not to ban AI, but to create verification barriers. Hallucination detection tools — like real-time quote checkers — are in development, but not yet widely adopted. Journal editors should require authors to explicitly declare the use of AI and submit references to human verification. In addition, universities may include training on AI hallucinations in their research ethics programs. For the individual researcher, the rule is simple: never rely blindly on a reference generated by AI; seeking the original article or the confirmed database is essential.

Frequently asked questions about AI hallucinations in science

Are AI tools reliable for academic research?

Not entirely. AI tools can generate false references with high confidence, so it is indispensable to verify each citation in reliable sources such as Scopus or Web of Science.

How to detect a false reference generated by AI?

Distrust citations whose title seems generic, with authors little known or who do not appear on traditional bases. Seek DOI or the journal page — if there is no such thing as hallucination.

Who is responsible for checking the sources before publication?

The main author is the final responsible, but reviewers and editors should also act as filters. The adoption of automatic detection tools can help you share this responsibility.