Automated image integrity tools carry a clear ethical responsibility: to surface suspected issues for human review, not to deliver verdicts. Responsible tools must minimize false alarms and protect researchers from unwarranted scrutiny. A flag is the beginning of a review process, never the end of one.
Why Does This Matter Now?
Image integrity screening is becoming standard across scientific publishing. The Science family of journals adopted Proofig AI across all six journals in 2024 to detect altered images before publication. As automated image integrity screening scales, so does the ethical obligation to get it right, both in what these tools detect and in how their outputs are interpreted.
The Committee on Publication Ethics (COPE) has long emphasized that integrity investigations must be fair, evidence-based, and presume good faith. COPE’s Core Practices state that journals should have clear procedures for handling allegations and should ensure that authors are given the opportunity to respond before any action is taken. This principle applies directly to automated screening: a tool’s flag is an allegation that requires context, not a conclusion that warrants punishment.
Understanding this distinction is essential for researchers who encounter automated screening, whether they are using a pre-submission image integrity tool themselves or responding to a publisher’s query about a flagged figure.
What Are False Positives, and Why Do They Matter?
A false positive in image screening occurs when an algorithm flags an image as potentially manipulated or duplicated when no actual problem exists. For researchers, the consequences are real: time spent responding to editorial queries, anxiety about reputational damage, delayed publication timelines, and in serious cases, formal investigation of work that is entirely sound.
False positives are not hypothetical. They are a known limitation of any pattern-matching system applied to complex scientific images. A peer-reviewed study on automated detection of image manipulations in the biomedical literature (Bucci, 2018, Cell Death & Disease) documented both the promise and the limitations of algorithmic approaches, including the challenge of distinguishing genuine artifacts from problematic alterations. COPE’s Retraction Guidelines further underscore that any process triggered by a flag, whether from a human reviewer or an automated tool, must be proportionate and fair.
This is why low false-positive rates are not just a technical performance metric. They are an ethical design priority. A tool that floods editors and authors with false alarms wastes researcher time, erodes trust in the screening process, and risks normalizing a culture of suspicion rather than support.
Do All Image Types Carry the Same False-Positive Risk?
No. Different scientific image categories have different structural properties that affect how detection algorithms behave. Microscopy images, with their complex textures and overlapping features, present different challenges than Western blots, which have more defined band structures.
Proofig AI is validated on large real-world datasets comprising hundreds of thousands of real microscopy and Western blot images and achieves very low false-positive rates across both image types. Internal validation shows that microscopy images carry a slightly higher false-positive rate than Western blots, which is consistent with the greater visual complexity of microscopy data. Both rates remain extremely low, reflecting the priority placed on minimizing false alarms during tool development.
These differences mean that researchers working primarily with microscopy images may encounter slightly different screening behavior than those working with blots. Understanding this helps set realistic expectations and reinforces the importance of reviewing flagged regions in context rather than treating every flag as equivalent.
The practical implication for researchers: a responsibly designed tool does not waste your time with constant false alarms. But when a flag does appear, it warrants review, not panic.
Most Image Issues Are Honest Mistakes
The ethical design of automated tools must account for a well-documented reality: the majority of image integrity issues are unintentional. Across the published literature and at the submission stage alike, a large share of flagged image problems are consistent with honest error rather than deliberate manipulation.
Image-related issues are common, particularly at the submission stage. According to data from experienced image screeners, an estimated 20 to 35% of submitted life-science manuscripts are flagged for image-related issues. Many of these reflect figure-preparation errors, accidental reuse of panels, or mistakes introduced by collaborators and students handling data across shared workflows.
A peer-reviewed study testing medical students and researchers at the University of Split found that researchers correctly identified only 32.4% of image duplications by eye, demonstrating the limits of unaided human review and explaining why errors persist even among careful teams.
This context shapes how tools should be built and used. Automated screening should operate on the assumption that most flagged issues reflect preventable errors, not deliberate misconduct. Even researchers who uphold the highest ethical standards may unknowingly submit problematic images because of mistakes introduced by team members or collaborators earlier in the workflow.
What Does Ethical Tool Design Look Like?
Responsible automated detection should follow principles that protect author rights while supporting the integrity of the scientific record. Based on COPE’s Core Practices and the practical realities of image screening, these principles include:
- Transparency. Researchers should be able to see what was flagged and understand why. A tool that produces opaque scores without showing the specific regions of concern fails the transparency test.
- Low false-positive rates as a design priority. Minimizing false alarms is not optional. It is a direct expression of respect for researchers’ time and reputation.
- Clear communication that flags are suspected issues. Every output should be framed as requiring human review. Language matters: “suspected duplication” is accurate, while “confirmed manipulation” is not the tool’s determination to make.
- The right to respond. Researchers must have the opportunity to provide context, explain legitimate reuse, or correct errors before any editorial or institutional action is taken.
- Responsible data handling. When evaluating any screening tool, confirm that manuscript data is processed only for the stated screening function and not repurposed, in line with the tool’s stated data practices.
These are not aspirational ideals. They are baseline requirements for any tool operating in this space. Proofig AI, for example, is designed to surface suspected issues for human review, not to assign intent or determine misconduct.
What Should Researchers Expect When They Encounter Screening?
Whether you are choosing a pre-submission screening tool or responding to a publisher’s automated flag, you should expect:
- A clear explanation of what specific image region was flagged and what type of issue was detected (duplication, manipulation, or other).
- The ability to review flagged regions yourself, ideally with visual overlays or side-by-side comparisons.
- A defined process for providing context, corrections, or explanations before any decision is made.
- Assurance that your manuscript data is processed for screening purposes and not reused for other functions.
- Proportionate follow-up that distinguishes between minor figure-preparation errors and patterns that warrant deeper investigation.
If any of these elements are missing, it is reasonable to ask for them. COPE’s guidance supports the principle that authors deserve fair, transparent processes at every stage.
Screening as a Safeguard, Not a Source of Suspicion
Automated image screening, used responsibly, protects researchers as much as it protects journals and institutions. Running a pre-submission check gives you the opportunity to catch and correct issues, whether your own or those introduced by collaborators, before they become editorial problems, delayed publications, or worse.
The goal is not to create suspicion. It is to support the integrity of your work. Understanding what any screening tool flags, why it flags it, and what the review process looks like allows you to engage with confidence rather than anxiety. The best tools are the ones that respect both the science and the scientist.
