How can we keep our trust on science in a world where research integrity is increasingly challenged? Experts are discussing different paths to ensure the continued integrity of the scientific process. One of them, Dr. Maura Hiney is an Adjunct Professor at University College Dublin Institute for Discovery and chair of the ALLEA Research Ethics and Integrity Council. She has been actively advancing research integrity policy and is a member of the National Research Integrity Forum. She will speak at the workshop ‘The future of European research: Integrity, freedom and trust’ on 10 June 2026 in the European Parliament, hosted by the Panel for the Future of Science and Technology (STOA).
With the rapid integration of AI into research, what new risks do you see emerging regarding scientific integrity and data reliability?
Maura Hiney: AI systems are now widely used in research, from assisting with hypothesis generation and experimental design to supporting statistical analysis of results, drafting manuscripts and reports, and helping scientific journals achieve faster, more efficient publication workflows. AI algorithms can learn from large datasets and once internalised, make decisions that are experientially or intuitively similar to those of humans. This means that, for the first time, computers are no longer merely executing detailed, pre-written instructions but can derive dynamic solutions to problems from patterns in data that humans may not perceive. This capability has significant benefits but also poses risks to research integrity.
Many AI algorithms are ‘black boxes’, even to their creators. This lack of transparency can make it difficult to understand how an AI system reaches its decisions or predictions, and whether they are reliable, reproducible, or even true. Poor-quality or flawed outputs can arise from context-induced algorithmic bias, perfunctory citation, confirmation bias, or hallucinations of fictitious references. This has serious implications for the integrity of the research record, which is becoming polluted with ‘AI slop’. In response, each of us, regardless of our technical background, must develop the ability to ask critical questions about the AI systems we encounter. We need to inquire about the data used to train the system and whether it can be trusted.
AI use is also vulnerable to questionable research practices and readily lends itself to intentional cheating, for example by using AI to generate fictitious data, falsify images, plagiarise text, or, in the worst cases, create entirely fictitious research and outputs. The risks to research integrity are obvious in these instances. However, there are more subtle risks when accountability for the use of these tools and their outputs becomes blurred. It is not always clear who can stand behind the judgements of such tools, or who is responsible for data privacy, security, and confidentiality. Without transparency, training, and clear policies, AI can weaken the trust that robust research assessment and research integrity are meant to uphold.
To what extent is scientific misconduct a problem of individual ethics vs a structural issue created by how careers and funding are evaluated in research systems?
Maura Hiney: Structural incentives clearly shape integrity outcomes. Research integrity and the assessment of research quality, rigour, reliability, and truth are closely connected. We tend to value what we measure, so how researchers and their work are rewarded and evaluated plays a vital role in whether research integrity and good practices are upheld within the research community. A growing body of empirical evidence across disciplines shows that a poor research culture within an institution influences researchers’ behaviours, amplifies perceived pressures and feelings of organisational injustice, and increases the likelihood of engaging in questionable research practices (QRPs). Research has also shown that traditional metrics correlate with methodological corner cutting. Not surprisingly, there is widespread concern among researchers that the current incentive system can distort research quality and fail to capture the real impacts of research.
Do you think the current academic publishing system, also known as “publish or perish”, indirectly encourages problematic practices?
Maura Hiney: The current research assessment model primarily relies on quick, comparable bibliometric indicators, such as publication counts and journal rankings, which are treated as ‘proxies’ for quality and excellence. This model is easy to use but does little to protect the integrity of research. Researchers often feel pressured to publish as much as possible, focusing on numbers rather than quality, in a ‘publish or perish’ culture perpetuated by the system, as institutions compete for funding and funding agencies struggle to secure the resources needed for fully qualitative evaluations. Unfortunately, this approach has several problems. Firstly, journal-level metrics don’t accurately reflect the quality or trustworthiness of individual papers or researchers and can be manipulated. Secondly, there is a strong correlation between journal rankings and the likelihood of retractions and misconduct. Thirdly, prioritising quantity over quality encourages questionable practices such as salami slicing, p-hacking, and data fabrication, which can undermine the overall reliability of research. Additionally, this system isn’t fair across all fields, as disciplines like the Humanities or the Social Sciences often emphasise books and monographs rather than articles, yet the same metrics are still used. Despite these problems, many institutions and funders continue to rely on these quantitative indicators to assess researchers.
If public trust in science continues to decline, what could be the long-term consequences for both scientific progress and democratic societies?
Maura Hiney: Being a researcher carries great responsibility. As Sheila Jasanoff of Harvard University notes, “There is hardly a step we take in the course of an ordinary day without […] relying on the judgment and virtue of anonymous experts. What have they done to deserve our trust?” Research integrity may be the basis for researchers to trust one another, but, equally importantly, it underpins society’s trust in research evidence and expertise. However, the challenge for most consumers of research evidence, whether policymakers or the public, is that they may not understand the methodologies or interpretations underlying the evidence. Reading about another high-profile misconduct case in the media makes it difficult for the public and policymakers to know whom to trust.
A disbelieving public is more susceptible to the scare tactics and misinformation of pressure groups and political actors, whose arguments are intended to discredit the research process and bolster their own cause. Sadly, we have seen this play out in many right-wing regimes around the world, where there has been significant political interference in what (and whom) can be researched and which outcomes are ‘acceptable’. The demonisation of vaccine research in the US and bans on gender research are good examples of this. This has led to ‘self-censorship’ by researchers, who are more likely to propose safe topics, resulting in emerging knowledge gaps in areas deemed off-limits by political paymasters. Attacks on academic freedom and truth are de facto attacks on democracy. Therefore, any actions that enhance public trust in research should be a key concern for all research organisations, and research integrity plays a central role in maintaining and supporting trust. This extends beyond the research process to include how research is disseminated and communicated, as well as how stakeholders are engaged in the research endeavour.
On a more practical level, over the past 25 years there has been substantial and growing public investment in European research across all spheres, including the humanities and social sciences. Much research is financed through taxation, so researchers depend on taxpayers to support their work. Where public trust is in decline, dissatisfaction may prompt Governments, who represent that public, to reduce investment or limit research funding to areas where they see potential economic impacts. Therefore, any disclosures of misconduct that come to light can have serious implications for the scale of (dis)continuing public investment and risk a country’s intellectual capacity.
What concrete changes should Europe prioritise to strengthen research integrity?
Maura Hiney: I believe the most important change we can make is to reform how researchers are evaluated for recruitment, funding, and promotion. Many international initiatives, including CoARA, DORA, SCOPE, and EviR, are re-examining research assessment norms to base assessment on qualitative evaluation and to avoid reliance on organisational and journal rankings. While these initiatives may focus on research funders, institutions, or governments, they all share a desire to shift towards more value-centric assessment. It is hoped that this reform will move the evaluation model towards one that assesses research on its own merits, recognises and rewards behaviours that strengthen research integrity, increases value and reduces waste in research funding, and, above all, recognises the diversity of contributions to, and careers in, research. On the downside, it is hard to change an assessment system that has served many senior researchers well in their careers and which is ingrained in the research ecosystem. A fair criticism of reform is that qualitative research assessment is burdensome for researchers, assessors, and institutions and is challenged by human variability. As a result, there is considerable interest in harnessing AI’s potential to facilitate this reform. But we have some way to go before we can be fully confident about the efficiency and consistency of AI in research assessment. There is still evidence of confirmation bias, as well as issues with inter-rater reliability and tool-dependent disagreement among different AI models. So, although the tools may appear to enhance efficiency, they struggle with nuanced contextual interpretation and consistency and may aggravate inherent bias built into the training data.
There is also the problem that when old metrics (e.g. impact factor, citation counts) are replaced with new ones (e.g. altmetrics, openness scores, impact case studies), researchers may still optimise for the metric rather than for the science. The risk is that researchers will begin to tailor their behaviour to score well with AI systems, potentially exacerbating superficial compliance at the expense of substantive quality and damaging the integrity of the research record.
Despite the risks of AI, it is here to stay, so we need to start thinking about how to mitigate its negative impacts on the integrity of research and research assessment. Most guidelines on the responsible use of AI in research centre on five activities. Firstly, ensure a human-in-the-loop by design, prioritising hybrid frameworks that integrate AI’s capabilities and scalability with human oversight and contextual judgement. The earlier in the research life cycle that good research practices and integrity are considered, the easier it will be for researchers to maintain and deliver high-quality research. Secondly, develop policies that improve transparency by mandating AI disclosure standards for journal publications and explainability in AI evaluation frameworks. Thirdly, it will be critical to build the capacity of editors and reviewers through training and accreditation for human evaluators working with AI, to embed integrity at every step. Fourthly, governance of AI use in research needs to be strengthened through structured guidance for responsible AI use that integrates the fundamentals of research integrity. Finally, Pandora’s Box is already open, and we must accept that we are not innocent bystanders, helpless in the face of seismic technological development. We need to embrace accountability, transparency, and human oversight to harness the potential of AI in research and preserve the integrity of that research.