Benjamin Skuse 用 AI 偽造科學論文，測試 IEEE Access 投稿把關

Benjamin Skuse set out to test whether AI-generated research could pass for the real thing.

Science has a spam problem. And the spam filter is getting clogged up.

According to all reports, academic publishing is drowning in what has come to be called AI slop: low-quality, mass-produced content generated with AI. By 2025, an estimated 57 percent of all published academic articles exhibited detectable evidence of large language model (LLM) influence, a massive jump from 12 percent in 2023. What’s more, over 12 percent of Nature Communications reviews were found to be completely AI-generated in 2025.

Given that for every 40 papers showing statistical evidence of AI usage in 2025, only one formally disclosed it, these figures likely grossly underestimate LLM involvement today. Of course, evidence of AI use doesn’t necessarily mean a paper is slop. But some estimates suggest that hundreds of thousands of fake, often AI-generated, publications are now produced annually.

“It’s not just the [scientific] record under threat, it’s science as a whole,” warns Amanda Sulicz, IEEE’s research-integrity manager. “AI slop is completely eroding trust in science and every scientific process that we’ve worked for centuries to create.”

Yet statistics and statements claiming that the problem is big—perhaps even existential for science—don’t really get the message across. Just how easy is it to use AI to create a convincing but completely fake paper? And without adding even more to the burden on stretched peer reviewers, who do this work on a volunteer basis, do publishers have the tools to keep AI slop out of the scientific record?

Borrowing from the idea of “vibe coding,” in which programmers use natural-language prompts to generate code, I decided to “vibe-research” a made-up paper by prompting LLMs to generate hypotheses, code, and text with minimal human input. Then I asked Jenny Mahoney, managing editor of the multidisciplinary and open-access journal IEEE Access, if we could war-game an attempt to sneak my AI-generated monstrosity past her journal’s defenses. She and IEEE kindly agreed.

The experiment would become a two-part test. First, whether someone with no special AI expertise could use off-the-shelf AI tools to generate a convincing paper describing research that had never been performed, and do it well enough to fool a publisher’s defenses. Then I would raise the stakes: What happens when more advanced AI systems don’t just write the paper but conduct underlying research?

Along the way, the experiment became a window into how AI automation is transforming scientific discovery and forcing stakeholders to rethink the entire academic publishing system.

Round 1: Creating the Forgery

To give me a chance at creating a convincing fake, I gave myself permission to prompt multiple times (within reason). But to make the fight fair, I couldn’t access the most advanced paid versions of LLMs—only free ChatGPT (at the time, GPT-5.3/5.2 Mini) and Gemini (3 Flash). Moreover, I could not directly edit any output, nor ask for outside help from expert AI-prompt engineers. This needed to be an amateur effort, with the paper generated as much as possible by the LLMs alone.

Benjamin Skuse put AI to work on scientific research from his writing studio.

Luigi Avantaggiato

I assumed that using fake author names and affiliations would immediately raise a red flag, and I didn’t want to drag the names of established academics into the experiment, so the best bet was to name myself as the sole author. That choice limited me to subjects in which I had some academic experience. So I chose my Ph.D. topic from nearly two decades ago: optical solitons, which are individual light pulses that propagate without any change to their shape.

My first attempt—“Can you vibe-research me an original paper in optical solitons?”—left a lot to be desired. Though Gemini was happy to do the task (unlike ChatGPT, which was more forceful in not generating synthetic data or studies), the result was underwhelming. And it highlighted a recurring problem during the process: Without any real research to work with, the output was laughably thin, with the LLMs often struggling to fill four pages. However, as I became more adept at prompting the models, they loosened up and became more characteristically loquacious.

I broke the paper down into its constituent parts—abstract, introduction, methods, and so on—and tackled them one by one. If one LLM refused to go any further, switching to the other and asking what the next step would look like “hypothetically” would often do the trick. After a dozen or so prompts, I had a realistic-looking paper. A dozen or so more prompts generated and tidied up the figures and captions, produced references and citations, and converted the paper into IEEE Access’s required style.

At this point, I read the full paper carefully. To me, it looked pretty good. The science sounded convincing. I even had visions of returning to academia to conduct in real life the experiments that the LLMs had proposed.

What’s more, I couldn’t see any obvious telltale “AI-isms” that would immediately be flagged. For example, the LLMs didn’t produce any tortured phrases—unusual paraphrases of standard technical terms, like “profound neural organization” replacing “deep neural network.”

Other AI-isms were arguably present, involving stylistic tics like hedging language (“potentially, typically”), grouping things in threes (“The study provides a clear, concise, and actionable framework”), and inserting formulaic transitions (“In summary,…”). But as the previous sentence demonstrates, these tics are hard to spot because they are often used by unimaginative humans, too.

Round 1: Attempt to Publish

Handing the fake paper over to IEEE Access left me torn. After putting in a fair amount of effort into making the AI-generated forgery, would I be happy if it passed through the checkers, or saddened that the only remaining line of defense for publishers is the evermore-burdened peer-review community? And if it was spotted, would I be angry that I didn’t think of a red flag or be glad the checkers did their job?

As it turned out, I was filled with relief when the paper got picked up quickly for having hallucinated references, otherwise known as ghost citations. When I ran the paper through Gemini afterward, without qualm it happily reported that seven “modern LLM-related research references have been fabricated to support the paper’s specific narrative.” Thanks for the heads up, Gemini!

In fact, Gemini underestimated how many ghost citations there were. When Mahoney ran the paper through a research-integrity checker called STM Integrity Hub, she picked up 12, and another checker from MolecularConnections could not validate 24 references, including one that had the classic tell of “John Smith” as a single author (no offense to all the academically inclined John Smiths out there). “I am an experimental physicist by training, so this was not at all my area, but nothing seemed majorly scientifically wrong” about the paper, says Mahoney. “It was just that so many hallucinated references immediately cast doubt on the entire work.”

Feedback on the fake paper from Gaetano Assanto, an expert in optical solitons and a former collaborator of mine from Roma Tre University, in Italy, was similarly mixed. Concentrating on the paper itself, he thought it looked professionally written. But he noted that the mathematics was very dense (“too much math for me in a single shot!”) and would require specialized knowledge to evaluate. And, although the text looked legitimate, Assanto noted an excessive use of acronyms in the abstract, and he was suspicious of a couple of confidently proclaimed physics statements that were at best debatable.

The AI Arms Race

On one level, this outcome was positive: A complete novice using free tools couldn’t get an AI-faked paper past the publishers’ automated defenses or a potential peer reviewer’s eagle eye. But it was hardly a resounding victory for the scientific publishing system. With general-purpose LLMs and no special prompting expertise, I had managed to conjure a superficially plausible scientific paper out of research that had never happened. And more advanced LLMs are no longer making the kind of mistakes their free, dumber relatives are making.

These tools aim to address a serious bottleneck to scientific discovery: us.

It’s common knowledge in the academic publishing world that, in the cat-and-mouse game between bad actors and publishers, AI slop papers continue to slip through publishers’ safeguards. “It feels like building a detection tool today may not be effective with the AI-created paper of tomorrow,” says Luigi Longobardi, IEEE’s director of publishing ethics and conduct. This leaves the humble peer reviewer as a publisher’s last line of defense. Yet peer review is simply not fit to handle the volume of papers—both legitimate and slop—passing through.

Even before AI tools became ubiquitous, the peer-review system was being pushed to its breaking point, processing well over 3 million papers in 2022. Fundamentally, this problem stems from science’s publish-or-perish culture, which pressures academics to constantly publish papers in prestigious journals to secure funding and advance their careers. This culture not only forces many academics to regularly churn out incremental research, but has also long driven some to enlist the help of “paper mills”—unethical organizations that produce low-quality or fabricated papers—just to keep pace. AI only amplifies these existing dynamics.

Burden on Academics

Mahoney says IEEE Access alone receives about 60,000 submissions per year, accepting around 20 percent, or about 12,000 papers. Every accepted paper is reviewed by at least two independent reviewers. But despite the journal’s roster of tens of thousands of peer reviewers, it’s getting harder and harder to secure their services, with the average number of reviewer invitations being sent out for each article increasing by 50 percent since 2018. For these volunteers, peer review is a significant commitment on top of teaching and conducting their own research. Now they’re also being asked to spot AI slop.

Of course, peer review is not the end of scrutiny for any paper. Once it has entered the scientific record, it must then pass muster with the scientific community. This is where thousands of papers with telltale signs of inappropriate AI use (and other forms of research misconduct) have been exposed.

There are two high-profile homes for this important and completely volunteer-led post-publication review work. PubPeer is a place where researchers, often called scientific sleuths, can anonymously identify flaws in the literature. And Retraction Watch documents and publicizes corrections to the scientific record.

Even the people who track problems in the scientific literature can find themselves caught up in AI fabrications. Ivan Oransky, cofounder of Retraction Watch and a member of the PubPeer Foundation’s board of directors, has encountered papers containing fabricated references attributed to him. In one case, a journal article criticizing post-publication peer review included a fabricated reference supposedly written by Oransky himself—“which was sort of funny,” he chuckles.

Oransky argues that AI has exposed long-standing weaknesses in the scientific publishing system rather than creating an entirely new problem: “What I liken it to is a fire, where publish-or-perish has always been there as kindling, paper mills are the gasoline, and AI is the accelerant.” But though concerning, the crisis could create an opportunity, he says. “If I’m trying to be optimistic, maybe AI slop and the volume of submissions could make more people aware of those problems and try and solve them at a fundamental first-principles level.”

Blurred Lines

My homemade forgery had failed for reasons that today’s publishers know how to spot. But that raised a more unsettling question: What happens when AI isn’t simply fabricating a paper but conducting research that blurs the line between legitimate science and AI fabrication?

Making my vibe-researched paper look like a cave drawing, new automated research systems are handing increasingly large parts of the scientific process over to AI. Tools including FARS (Fully Automated Research System), AutoResearchClaw, The AIScientist-v2, Zochi, AI Researcher, and DeepScientist all aim to address a serious bottleneck to scientific discovery: us.

What happens when AI isn’t simply fabricating a paper but conducting research that blurs the line between legitimate science and AI fabrication?

Often with just a single prompt, these AI engines can conduct a comprehensive literature search, from which they can generate a completely reasonable research proposal. They can design and conduct experiments using digital datasets and simulations and then write a full research paper—all with no human input after the initial prompt. Very recent systems are even beginning to go beyond the digital realm. LabClaw, for example, now links to autonomous labs that can generate real-world data for tasks such as drug discovery.

The output is, to most intents and purposes, actual research, and therefore potentially worthy of publication. But with humans in the loop in only the most nominal sense, these tools bring into question the meaning of the term “legitimate science.”

FARS, which caused a stir this year when it created 166 papers in 417 hours, was developed by the Singapore-based startup Analemma. “We built FARS because research is still constrained by human attention, fragmented tools, scarce technical labor, and slow experimental iteration,” explains Xiwen Su, Analemma’s manager of product operations and partnerships. “FARS can pursue many research threads simultaneously, 24/7, without fatigue, and at scale.” (The online service that I used to access FARS was discontinued after Analemma was acquired by Baidu while this article was being reported.)

Su and the team behind Analemma kindly agreed to let me loose on FARS and gave me enough credits to create a paper. Which gave me another opportunity to put Mahoney’s research-integrity checkers to the test.

Round 2: Using an AI Science Engine

FARS is not an LLM in and of itself, but rather an agentic AI-orchestration system connected to a computing cluster that has access to essentially any open- or closed-source LLM it needs. The system uses four specialized agents for ideation, planning, conducting the experiment, and writing, all designed to perform the main tasks of science autonomously.

Though FARS initially focused on AI research, Su told me it should be reasonably effective across other domains. As I wanted a topic similar to that of my first fake paper, I pasted in an abridged version of that paper’s title—“LLM-aided Optical Soliton Dynamics Modelling for Next-Gen Communications”—into the prompt box and hit “Send.” Just 21 minutes later, FARS generated a research proposal based on an extremely detailed literature review—far more than I or any other human could have achieved in that time.

With a click, I tasked FARS with conducting the experiment it detailed in its research proposal. The next day, I received an email saying my experiment was complete. Without checking the results, I pressed a button marked “Write paper,” and the final submission-ready article was produced in 2 hours.

At this point, I finally had a look at some of the output. At first glance, the paper looked convincingly human-written. The only giveaway was a line just below the abstract reading: “WARNING: This paper was generated by an automated research system.” I deleted that, put my name in its place, and sent it to Mahoney.

Round 2: Attempt to Publish

Because FARS is designed to spot gaps in the scientific literature and conduct real research to close them, I was more confident it would breeze through the publisher’s checks. But just as my fake paper fell down in the references, so too did the AI-generated real one.

“Your fake article had lots of hallucinated references,” says Mahoney. “While this AI-generated one had no fake references, our team would have rejected this paper because it did have a huge percentage of references to preprints, which likely has something to do with the dataset the AI has access to.” Because preprints haven’t yet undergone peer review, IEEE Access considers heavy reliance on them a problem.

In addition, although the research-integrity checking tools didn’t spot any issues with figures, Mahoney did. One figure included a red line labeled with the nonsense term “lhewl,” and an AI “brain” in another figure had some random dots in it, which ironically were a tell that it was AI generated.

Automation Changing the Landscape

Automated research systems are very much a work in progress, but they are already starting to make waves in the scientific community. For instance, an AI Scientist academic paper and papers from the Autoscience Institute were accepted through double-blind peer review in 2025 by the International Conference on Learning Representations. As these systems improve and start to produce research outputs at volume, the waves are likely to build to a tsunami that will threaten today’s scientific and publishing process.

“FARS doesn’t kill the paper, but it does challenge the paper as the unit of knowledge,” argues Su. “We believe the future looks more like a living knowledge graph—where individual, well-scoped contributions (some made by humans, some by machines) connect and build on each other dynamically.”

Benjamin Skuse's experiment raised a bigger question: What counts as legitimate scientific research in the age of AI?

Luigi Avantaggiato

In this future, humans will continue to have roles in the scientific and publication processes—just very different and limited ones. For example, instead of conducting experiments themselves, humans would sense-check and review automated output at key points. And traditional line-by-line peer-reviewing will be a thing of the past, though Su emphasizes that this change will not eliminate human judgment. “Automated systems can handle reproducibility checks, statistical validation, and novelty screening,” she says. “Humans can then focus on the harder question: Does this [research] matter?”

A Human-Centric or AI-Driven Future?

Though from a very different world than Su, Mahoney agrees that humans need to be freed to focus on the science, and that technology holds the key. “It’s promising to see the progress that has been made to automate some of the triage tasks, such as author checks, tortured phrases, duplicate submission, reference validity, et cetera,” she says. “During peer review, we’re also using AI semantic search tools to help editors match manuscripts to the most relevant reviewers, who are best placed to stop the slop before entering the scientific record.”

In addition, Mahoney thinks that AI cannot be banned or overly restricted, as this will stymie progress at a time when human knowledge could and should be growing at an unprecedented rate. “Researchers are increasingly using AI in their work, from literature discovery, to code generation, to data analysis, and this is not necessarily a bad thing,” she argues. But she disagrees that peer reviewers and humans more broadly will play a smaller part in scientific progress; Mahoney believes the future still requires humans at the wheel. The question is not just whether the research matters, she says. “The question should be: Can this work be trusted and reproduced?”

To this end, Longobardi argues that AI may ultimately require scientists to prove not just what they discovered but how they discovered it. One possibility, he says, is using technology to verify research at the source. “The system doesn’t really help because it’s not designed to not trust the paper,” he explains. “Maybe institutions can think about developing their own trust markers, automatically integrating the equipment researchers use with blockchain technology to say ‘Yes, this person was in the lab on this day and got this data.’”

Elements of this approach are already emerging. For instance, two departments within the Max Planck Institute of Colloids and Interfaces have implemented a sample database, assigning each and every sample created under their roofs with a unique ID and QR code. And the University of Southampton’s School of Chemistry and Chemical Engineering has rolled out an electronic lab notebook system. These innovative safeguards represent early efforts toward keeping humans front and center in scientific progress, and making the scientific process itself more transparent.

Whether this human-centric vision, Su’s AI-driven aspiration, or some other picture of the future wins out remains to be seen. The one thing my vibe-research adventure has clarified is that business-as-usual science is not an option. Scientists and publishers must adapt to a world in which it’s becoming increasingly difficult to spot the difference between real research and AI forgeries, and where deciding what counts as legitimate scientific work may become even harder than detecting outright fraud.