AI and Image Manipulation in Research: Safeguarding Scientific Integrity

AI and Image Manipulation in Research: Risks, Detection, and How to Safeguard Scientific Integrity

Introduction

Artificial intelligence (AI) has rapidly become embedded in virtually every stage of the research process. From analysing complex datasets to segmenting medical images and automating statistical pipelines, AI can dramatically speed up scientific workflows and reveal patterns that would otherwise be missed. Yet, alongside these benefits, AI has also introduced a powerful new avenue for image manipulation in scientific publications.

Figures and images are not decorative extras in research papers; they are often central pieces of evidence. Microscopy images show cellular changes, blots reflect protein expression, medical scans illustrate pathology, and simulation outputs visualise complex physical systems. When these images are accurate and appropriately processed, they help readers evaluate the robustness of a study. When they are manipulated—especially with sophisticated AI tools—they can fundamentally distort the scientific record.

The recent growth of AI-based editing and image generation tools has made it significantly easier to enhance, alter, or fabricate research images. Minor adjustments such as noise reduction or contrast enhancement can be legitimate and even necessary; however, the same techniques can be pushed into ethically unacceptable territory when they remove real data, create artificial structures, or mislead readers about what experiments actually showed.

This article examines the dual role of AI in scientific imaging. It explores how AI can ethically improve image quality and support analysis, but also how it can be misused to fabricate results and mislead the scientific community. It then discusses the impact of AI-driven image manipulation on research integrity, surveys AI-powered approaches to detecting fraud, and outlines concrete steps that researchers, journals, and institutions can take to prevent and respond to AI-assisted image misconduct.

The Role of Images in Scientific Research

Images play a particularly important role in many disciplines, including biology, medicine, chemistry, physics, materials science, and astronomy. Common examples include:

• Microscopy images showing cells, tissues, or subcellular structures.
• Medical imaging such as X-rays, MRI, CT, or ultrasound scans.
• Western blots, gels, and other assay readouts used to quantify proteins, DNA, or RNA.
• Simulation and modelling outputs that depict fluid flows, molecular dynamics, or climate models.
• Astronomical images capturing galaxies, exoplanets, or cosmic background radiation.

These images do more than illustrate a story—they support claims and often underpin quantitative analyses. Consequently, altering them inappropriately can change the apparent outcomes of experiments and skew conclusions, even if the accompanying text remains unchanged. This is why most publishers now provide explicit guidelines on what kinds of image processing are acceptable—for example, brightness and contrast adjustments applied uniformly—and which practices, such as splicing lanes without annotation or selectively erasing features, constitute misconduct.

AI in Image Processing: Ethical and Unethical Uses

AI-driven tools are used in a growing range of image-related tasks. The key distinction is not whether AI is used but how it is used and whether the underlying data remain faithful representations of reality.

Ethical Uses of AI in Scientific Imaging

When applied transparently and within agreed guidelines, AI can greatly enhance the quality and interpretability of research images. Legitimate applications include:

• Resolution enhancement: Deep-learning models can upscale low-resolution images, revealing details that might otherwise be difficult to see, especially in low-light or low-dose imaging where raw data are noisy.
• Noise reduction and artifact removal: AI can filter out random noise from microscopy, astronomical, or medical images without altering underlying structures, provided the process is validated and documented.
• Automated segmentation and quantification: AI-based image analysis can identify cell boundaries, lesions, or features in large image sets, enabling consistent and reproducible measurements at scale.
• Data visualisation: AI can help generate clear, structured representations of complex multidimensional datasets, for example by highlighting relevant regions or generating heatmaps for statistical results.

In all of these cases, ethical practice requires that the AI pipeline be transparent, validated, and disclosed. Authors should be able to show how processed images relate to the raw data and explain which adjustments were made and why.

Unethical Uses: AI-Facilitated Image Fabrication and Manipulation

The same capabilities that make AI useful can be exploited for misconduct. Unethical uses of AI in research images include:

• Altering experimental results: Using AI-based editing to remove blemishes, bands, or data points that contradict a hypothesis, or to intensify signals to make effects look stronger than they are.
• AI-generated “deepfake” scientific images: Creating entirely artificial microscopy or imaging data that never came from real experiments, then presenting them as genuine results.
• Duplicating and reusing images with subtle modifications: Copying an image from another study—or another experiment within the same study—and using AI tools to flip, crop, adjust colours, or add synthetic variation so that it appears to show a different condition.
• Selective editing and cropping: Removing inconvenient portions of an image (for example, failed experiments or inconsistent lanes in a blot) while leaving the rest intact, misleading readers about variability or background signals.

As AI tools become easier to use and more powerful, the technical barrier to such manipulation is falling. This has contributed to a noticeable rise in image-related concerns and retractions in the literature, prompting journals to invest in more sophisticated screening tools.

The Impact of AI Image Manipulation on Scientific Integrity

Loss of Trust in Research

Science depends on trust: trust that methods are reported honestly, that data are not fabricated, and that figures accurately represent experimental results. When AI is used to manipulate images, it directly undermines this trust. Even a small number of high-profile fraud cases can create widespread suspicion, especially in sensitive areas such as clinical trials or pharmaceutical development.

Misguided Research and Wasted Resources

Fraudulent images are not only unethical; they are also harmful to progress. If other scientists build their own experiments on fabricated data, entire lines of inquiry can be distorted. Time, funding, and effort may be invested in trying to replicate results that were never real, delaying genuine advances and crowding out more promising work.

Retractions, Sanctions, and Damaged Careers

When manipulated images are discovered after publication, journals may retract the affected papers. Retractions are publicly visible and can have long-term consequences:

• Authors may lose research funding, career opportunities, or academic positions.
• Co-authors and institutions can suffer reputational damage, even if they were not directly involved in the misconduct.
• In extreme cases, legal or regulatory bodies may become involved, particularly in fields relating to patient safety or environmental risk.

Damage to Public Confidence in Science

In an era of rapid communication and social media, cases of scientific fraud quickly reach the public. When misconduct involves AI-manipulated images in areas such as cancer research or vaccine development, it can feed conspiracy theories, fuel scepticism, and make it harder for policymakers and clinicians to rely on scientific advice. Protecting image integrity is therefore not only an internal academic issue; it is also a matter of public trust.

How AI Is Used to Detect Image Manipulation

Fortunately, AI is not only part of the problem—it is also part of the solution. The same techniques that enable sophisticated image editing can be employed to identify signs of tampering and support editors and reviewers in safeguarding the literature.

AI-Powered Image Forensics

AI-based forensic tools can analyse images for subtle irregularities that may indicate manipulation. These systems can detect:

• Inconsistent pixel patterns that arise when elements from different images are combined.
• Lighting and shading anomalies that suggest objects were artificially inserted or removed.
• Cloning and duplication artifacts where regions of an image have been copied and pasted elsewhere.

These tools can operate at a scale that would be impossible for human reviewers alone, scanning large numbers of submissions and flagging suspicious figures for further examination.

Image Plagiarism and Reuse Detection

Just as plagiarism-detection services compare text against large databases, specialised tools can compare research images against repositories of previously published figures. They can identify:

• Reused images that appear in multiple papers but are presented as distinct experiments.
• Cropped, rotated, or colour-adjusted versions of the same image used in different contexts.

This helps editors spot paper mills or serial offenders who recycle the same visual data across many publications.

Pattern Recognition in Domain-Specific Images

Machine learning models trained on domain-specific datasets—such as histology slides, gel images, or astronomical photos—can learn what “normal” patterns look like. They can then detect implausible structures or textures that might indicate artificial generation or manipulation.

Blockchain and Provenance Tracking

Some institutions and consortia are experimenting with blockchain-based systems to record and verify the provenance of research images. By assigning a unique cryptographic signature to raw images at the time of acquisition and storing that signature in a distributed ledger, it becomes possible to confirm whether a published image corresponds to the original data or has been altered.

Hybrid Human–AI Review Models

Even the best AI tools cannot fully replace expert judgement. Many journals are moving toward hybrid workflows in which:

• AI systems pre-screen images and generate reports on potential anomalies.
• Editors and experienced reviewers evaluate the flagged images in context, checking against raw data and the study’s narrative.

This combination allows for efficient screening without abdicating human responsibility for final decisions.

Preventing AI Image Manipulation: Policies and Best Practices

Detection is important, but prevention is even better. A robust response to AI-assisted image manipulation requires coordinated action from researchers, institutions, funders, and publishers.

Establish Clear Ethical Guidelines

Universities, research institutes, and journals should publish explicit policies on acceptable and unacceptable image processing. These policies should distinguish between:

• Permitted adjustments such as uniform brightness/contrast changes or minor cropping for clarity.
• Prohibited manipulations including deleting or inserting features, splicing images without annotation, or using AI to generate synthetic data presented as real.
• Disclosure requirements when AI-based tools (for enhancement or analysis) have been used.

Integrate Mandatory AI-Based Image Screening

Journals should incorporate AI-driven image analysis into their routine submission checks, particularly in fields where image-based evidence is central. This can catch many problems before articles reach peer review or publication.

Require Raw Data and Original Files

To enable verification, journals can require that authors submit raw image files (for example, original microscopy or imaging data) along with processed figures. Editors and reviewers can then:

• Check that published figures accurately reflect the originals.
• Confirm that any AI-based processing is transparent and justified.

Promote Open Data and Reproducibility

Open data practices—where raw images, analysis scripts, and metadata are shared in trusted repositories—make it easier for other researchers to reproduce image-based findings and to detect potential issues after publication. Transparency acts as a powerful deterrent to misconduct.

Train Researchers in Responsible AI Use

Early-career researchers may not fully appreciate the ethical boundaries of AI-based image manipulation. Institutions should offer training that covers:

• The difference between legitimate enhancement and fraudulent alteration.
• The risks associated with AI-generated images and deepfakes.
• Best practices for documenting and disclosing image-processing workflows.

Strengthen Sanctions for Misconduct

To deter AI-assisted image fraud, there must be real consequences when it occurs. Possible responses include:

• Public retraction of affected papers with clear explanations.
• Temporary or permanent bans on submission for authors found guilty of serious manipulation.
• Reporting to employers, funders, and, where appropriate, regulatory bodies.

The Role of Human Oversight and Independent Checking

Ultimately, AI alone cannot guarantee research integrity. Humans must remain responsible for designing experiments, interpreting data, and ensuring that images and figures faithfully reflect reality. This includes:

• Supervisors carefully reviewing figures produced by students and early-career researchers.
• Co-authors scrutinising images for inconsistencies before submission.
• Editors and reviewers asking for clarification or raw data when image processing appears excessive or unclear.

Many researchers also choose to have their manuscripts and figure legends reviewed by independent, human proofreaders and editors before submission. Unlike AI rewriting tools, which can increase similarity scores or inadvertently alter meaning, professional academic proofreading focuses on clarity, consistency, and style while leaving the underlying data and images unchanged—an important safeguard in an environment of growing scrutiny around AI use.

Conclusion

AI has brought remarkable advances to scientific imaging, enabling clearer pictures, faster analysis, and more efficient workflows. But it has also opened the door to new forms of image-based misconduct, from subtle manipulations to fully synthetic “deepfake” results. These practices threaten not only individual studies but the credibility of the scientific enterprise as a whole.

To respond effectively, the research community must treat AI as both a tool and a risk factor. AI-based forensic analysis, plagiarism detection for images, pattern-recognition models, and blockchain provenance tracking all have important roles to play in detecting manipulation. At the same time, robust ethical guidelines, researcher education, mandatory raw data submission, open data practices, and meaningful sanctions are essential for prevention.

The future of trustworthy science will depend on a multi-layered, hybrid approach: AI will be used to screen, support, and flag potential problems, but humans will remain responsible for final judgements and ethical oversight. By combining responsible AI deployment with strong human review—and by avoiding risky shortcuts such as AI rewriting in favour of transparent, human-centered support such as expert academic proofreading—the research community can harness AI’s strengths while protecting the integrity of the scientific record for generations to come.