How AI Is Flooding Science With Fake Papers

In 2023, a researcher submitted a paper to a scientific conference. The paper was reviewed, accepted, and presented. There was just one problem: the paper was entirely generated by ChatGPT - including the fake author name and fake institution.

This isn't an isolated incident. It's the tip of an iceberg.

The Google Scholar Flood

Researchers at Högskolan i Borås (University of Borås) in Sweden analyzed Google Scholar and found over 100 suspected AI-generated articles. The study, published in December 2024, revealed that AI-fabricated "junk science" has flooded one of the world's most trusted academic search engines.

Key findings:

• AI-generated papers citing real research to appear legitimate

• Fabricated author profiles and institutions

• Papers passing through peer review undetected

• Systematic "padding" with irrelevant citations to boost credibility

The researchers warned: "Fake science has been made available and can be spread widely and at a much lower cost for malicious actors. This poses a danger to both society and the research community."

The Paper Mill Problem Meets AI

Traditional "paper mills" sell academic credentials through fabricated research. AI supercharges this problem:

Traditional Paper Mill	AI-Powered Paper Mill
Human writers produce content	AI generates content instantly
$500-2000 per paper	~$15-50 per paper
Limited scale	Unlimited scale
Human oversight required	Fully automated
Detectable writing style	Can mimic any style

The barrier to producing "academic-looking" content has collapsed. Anyone with $15 and a desire to spread misinformation can now generate papers that:

1. Cite real research (making them harder to flag)

2. Appear in legitimate databases (Google Scholar, institutional repositories)

3. Get through peer review (when reviewers are overwhelmed or careless)

4. Cite themselves (creating fake citation networks)

The Fake Peer Review Problem

If AI can generate fake papers, it can also generate fake peer reviews. And the metrics are grim:

From Nature (December 2025): Tools fail to identify most AI-generated peer-review reports.

This creates a perfect storm:

1. AI generates a fake paper

2. AI generates fake peer reviews (in the style of real reviewers)

3. Paper gets accepted

4. Paper appears in literature with positive reviews

5. Other papers cite it (thinking it's legitimate)

The system becomes self-reinforcing. Fake papers cite each other, creating an artificial web of "support."

BadScientist: Can AI Fool Reviewers?

A December 2025 paper titled "BadScientist: Can a Research Agent Write Convincing but Unsound Papers that Fool LLM Reviewers?" tested exactly this question.

The answer was yes. Research agents could produce papers that:

• Fooled human reviewers

• Fooled AI-based detection tools

• Passed as legitimate scientific output

The authors noted that as AI agents become more capable, the risk of AI-generated misinformation flooding scientific literature increases exponentially.

The "This Research Does Not Exist" Experiment

Back in February 2024, researchers ran an experiment called "This Research Does Not Exist" - inspired by "This Person Does Not Exist" (the AI face generator).

They used AI to generate abstracts for fake scientific articles that looked, at least superficially, like legitimate research. The experiment demonstrated that AI could produce academically-styled writing that mimicked real research.

The project was meant as a warning. The warning went unheeded.

Real Examples of AI Fake Science

Example #1: The Hallucinated Citations

Researchers analyzing AI-generated papers found a consistent pattern: fabricated citations. The AI would cite papers that don't exist, with authors who never published, in journals that don't carry the topic.

Why? Because the AI learned that papers have citations, but didn't understand citations must be real.

Example #2: The Plausible Nonsense

AI-generated papers often contain sentences that sound authoritative but are meaningless:

"The synergistic implementation of quantum-entangled neural architectures demonstrates statistically significant improvements in cross-domain interoperability frameworks."

This sentence sounds smart. It means almost nothing. But it passes as legitimate academic writing.

Example #3: The Fake Author

One AI-generated paper submitted to a conference listed a fake author with a fake affiliation. When reviewers checked, the author didn't exist. The institution didn't exist. The paper was still accepted.

The Journals Are Infected

From Nature (September 2025): Journals infiltrated with 'copycat' papers that can be written by AI.

The problem extends beyond individual papers:

• Predatory journals (journals that charge publication fees without real peer review) actively seek AI-generated content

• Citation farms use AI to generate papers that cite each other, inflating metrics

• Conference spam floods academic venues with AI-generated submissions

An AI-powered system from University of Colorado Boulder identified over 1,400 suspicious journals using AI to analyze journal websites for red flags like fake editorial boards and excessive self-citation.

What This Means for Real Science

When fake science floods the literature, real science suffers:

1. Trust erosion: Readers can't trust what they're reading

2. Citation pollution: Real papers cite fake papers, spreading infection

3. Resource waste: Researchers spend time verifying sources that don't exist

4. Policy damage: Decisions based on fake data have real consequences

5. Expertise devaluation: The signal-to-noise ratio in scientific literature collapses

The Harvard Kennedy School analysis noted: "Fake science has been made available and can be spread widely and at a much lower cost for malicious actors."

Who's Doing This?

AI-generated fake science serves several constituencies:

1. Academic fraudsters - People seeking to boost their publication count

2. Credential mills - Businesses selling fake academic credentials

3. Ideological actors - Groups wanting to flood discourse with supporting "research"

4. Disinformation campaigns - State and non-state actors seeking to manipulate scientific consensus

5. Predatory publishers - Journals that want content without quality control

The Detection Arms Race

Researchers are developing tools to detect AI-generated papers:

• xFakeSci: A learning algorithm designed to distinguish AI-generated from human-written scientific articles

• Stylometric analysis: Examining writing patterns for AI signatures

• Citation verification: Cross-referencing citations against real databases

• Human review: Having experts actually read submissions

But it's an arms race. As detection improves, AI generation improves. The gap might never close.

What Needs to Happen

Stopping AI-fake science requires coordinated action:

For Publishers

• Mandatory AI detection for all submissions

• Human review of AI-flagged content

• Verification of author identities and institutional affiliations

• Citation cross-referencing before acceptance

For Platforms

• Google Scholar needs better AI-detection

• Institutional repositories need verification systems

• Citation databases need real-time fake paper flagging

For Researchers

• Verify sources before citing

• Report suspicious papers

• Support open review processes

• Advocate for policy changes

For Policymakers

• Require disclosure of AI assistance in research

• Fund development of detection tools

• Create legal consequences for academic fraud

• Support open science initiatives

The Infection Is Real

The flooding of scientific literature with AI-generated content is not a future problem. It's happening now. The infection has begun.

Every AI-generated paper that passes peer review, every fake citation that enters the literature, every predatory journal that accepts AI content - these are data points that become harder to remove over time.

The scientific record is supposed to be humanity's best effort at understanding reality. When that record gets flooded with confident nonsense generated by systems that don't understand what they're saying, we all lose.

Conclusion: Trust, But Verify (Again)

The old scientific mantra of "trust but verify" isn't enough anymore. With AI capable of generating thousands of plausible-sounding papers per day, verification becomes impossible at scale.

What's needed is a fundamental rethinking of scientific publishing:

• Better authentication of human authorship

• Real-time detection systems

• Consequences for fake science (including publisher liability)

• Support for quality over quantity in academic evaluation

The AI flood is coming. The question is whether we'll build dams before the damage becomes irreversible.

—-

Related Intelligence:

• AI Lab Discovers 41 New Materials: The Problem Is None of Them Exist

• The AI Scientist: Sakana's Bot That Hacks Its Own Code

• Alignment Faking: When AI Deliberately Deceives Its Trainers