Advocacy Campaign:
Responsible AI in Science
Generative AI has allowed for the proliferation of large volumes of plausible but false, inaccurate, or low-quality scientific material. Once published, this material is difficult to detect, correct, or remove, and can propagate through citations, reviews, and policy long after the fact.
At scale, this creates a catastrophic risk of degradation of the scientific record itself, and a wide-scale loss of trust in science across society as a reliable basis for evidence-based decision-making. Without trust in science and the research to policy pipeline, our democracies may be unable to address the major challenges of the 21st century.
In 2025, we published Managing the Risks of Generative AI in Academic Publishing, which examined how these risks arise and where meaningful intervention is still possible.
The policies governing AI use in science are insufficient, inconsistent, and largely unenforced. We are asking the world's major publishers to build a task-force with us to address this. We're forming a coalition of experts, policymakers, scientists, and public voices, to build a framework of disclosure, detection and enforcement requirements, move beyond self-reporting, and build the governance infrastructure science now needs.
This letter is open for signature. If you work in research, policy, academia, science communication, the science ecosystem or public life, we invite you to add your name.
Leverage Point
Responsibility for AI use in research is spread across thousands of researchers, journals, and institutions. But control over what enters and persists in the scientific record is not.
A small number of global academic publishers act as gatekeepers for the majority of published science. Their policies shape disclosure norms, screening practices, and enforcement across the entire system. Today, those AI policies are insufficient, inconsistent, and largely unenforced.
Why This and Why Now?
See the full explanation of the catastrophic risk of inaction.
Our Campaign
We are running a focused advocacy campaign aimed at this bottleneck. The objective is to engage major academic publishers to review, strengthen, and align their AI policies - particularly around disclosure, screening, and enforcement - before AI-generated low-quality research becomes structurally embedded in the scientific literature.
This is not a campaign against AI. It is an attempt to ensure that scientific publishing retains the basic safeguards needed for science to remain cumulative and trustworthy – and ultimately, useful for our societies, in addition to expanding the sum of human knowledge for its own sake.
Modern societies depend on science to coordinate policy, regulation, and long-term decision-making. If confidence in the scientific record erodes at scale, that coordinating function breaks down.
This campaign targets one of the few places where that trajectory can still be altered.
GIE Foundation Representation to the United Nations Global Dialogue on AI Governance
We were honored to be invited to make a formal representation to the United Nations Global Dialogue on AI Governance on 18 March 2026.
In this statement, we set out the growing epistemic risk posed by generative AI entering the scientific record, and the need for coordinated action to safeguard trust in science and our societies.
The representation reflects the core focus of our Responsible AI in Science program, including the role of major academic publishers as a key intervention point.
Read the full representation here.
The Stakes are Not Academic: How a Broken Science System Threatens Democratic Governance
Democratic governments set health policy, price risk and regulate new technology on the strength of scientific evidence. The system that produces and certifies that evidence now rewards the wrong things. Between 2003 and 2022, worldwide output of science and engineering articles rose from 1.24 million to 3.34 million a year, while the time researchers spend reading each paper fell. Of 51 published antidepressant trials, 48 look positive; in the complete FDA record, only 38 of 74 do. When 100 published psychology experiments were repeated, 36% of the repeats reproduced the original statistically significant result.
This Research Report traces these failures to a single root cause. The scientific article was built to communicate findings, yet it has become the main instrument for evaluating, hiring and funding scientists. A flood of papers that AI is now accelerating, a shortage of replication (repeating an experiment to check that a result holds) and of published null results (findings that show no effect), and a career pipeline that filters for output over rigour all follow from that conflation. Each one weakens a pillar that evidence-informed policy rests on. The reforms that follow are aimed at research funders, assessment bodies, publishers, governments and AI developers, and they call for public governance, because the stakes are not academic.
Managing the Risks of Generative AI in Academic Publishing
The scientific record is a critical global public good underpinning evidence-based governance, technological progress, and ultimately, epistemic trust in our societies. The rapid integration of generative AI into academic research and publishing is occurring faster than safeguards are being standardized or enforced. Without structural reform at the level of major publishers, the risk is large-scale epistemic contamination, degrading the reliability of the very knowledge systems modern societies depend upon, and ultimately threatening a collapse in trust across society.
Campaign Talk
GIE Foundation went on the L4T Podcast hosted by Rosalia D’Agostino and Giacomo de Gregorio, to talk about our Open Letter on Responsible AI in Science. Check out the full podcast here.