Test shows AI-generated scientific papers can bypass IEEE's review process
Journalist and mathematician Benjamin Skuse tried to push AI-generated scientific papers through IEEE's review process. Human review and integrity-checking tools, according to the source, are so far keeping up with freely available AI tools, but more advanced LLMs are harder to detect.
Benjamin Skuse, a freelance journalist with a PhD in applied mathematics, described an attempt to generate a scientific paper using AI and push it through IEEE's review process. According to the account in IEEE Spectrum, in one of the experiments he merely gave instructions to run the experiment and write the paper, made no further edits to the text, and submitted it in that form to the publication process.
IEEE combines human review with research integrity-checking programs. According to the source, this combination is so far keeping pace in a cat-and-mouse game with freely available text-generation tools, because those tools still make recognizable errors. However, according to the same source, more advanced language models no longer make these errors, which makes detection harder and creates a race between increasingly convincing AI content and the systems meant to detect it.
According to the article, Wikipedia faces a similar problem. The volunteer initiative WikiProject AI Cleanup, described by one of its founding members, Ellia Lebleu, tries to detect and delete AI-generated content, but has fewer people and tools at its disposal than commercial publishers.
The source article further explores what this development means for trust in the scientific record and shared reality in general; for details on these considerations, see the source article.
Why it matters
It shows that even established review processes and checking tools are not a reliable guarantee against more advanced AI models that published text is truly the result of human work or verified research. For institutions that rely on peer-reviewed content or community-built sources like Wikipedia, this means a growing risk that unchecked AI content will make its way into material considered trustworthy.
Two audiences, two different impacts
What this means
For individuals
Readers and professionals who rely on peer-reviewed publications or Wikipedia when making decisions should bear in mind that even such vetted content may be AI-generated without human editing, and where possible should verify key facts against other sources.
For a business
Organizations running publishing platforms, review processes, or community knowledge bases face the risk that more advanced language models will bypass current integrity checks; according to the source, this is forcing them to strengthen detection tools in a race against increasingly convincing AI content.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.