How we use AI responsibly
Artificial intelligence (AI) is transforming how people and organisations work. As an organisation dedicated to integrity and good governance, we at the Basel Institute on Governance hold our own use of AI to the same standards we promote. We see many opportunities in AI to strengthen our anti-corruption mission. We also take its risks seriously.
Our fundamental values – integrity, independence, accountability, transparency and professionalism – give us solid ground from which to assess AI’s impact, opportunities and risks. They guide how we use AI but ultimately keep people in charge of our work.
Our safeguards
Used well, AI helps us work more efficiently, improve quality and share our knowledge more widely. But we know it also comes with risk, which we manage through a robust policy, guidelines, training, clear responsibilities and approved tools. These are structured around four interconnected capabilities: the “4Ds” below, adapted from the AI Fluency Framework of Rick Dakan and Joseph Feller. The framework also distinguishes three forms of AI use – automation, augmentation and agency – which shape how we delegate tasks.
Delegation: what to entrust to AI
We think carefully about which tasks we can automate with AI; where AI can augment our work while we stay in charge, such as drafting or analysis; and which, if any, can be handed to AI agents acting with some autonomy. Decisions always stay with people.
Description: clear instructions, minimum sharing of information
We give AI tools clear instructions and only the information a task needs. Non-public work information is used only in approved tools, through Institute accounts, and never where a contract with a funder or partner rules it out. Particularly sensitive or confidential information never goes into any AI tool. Simple tasks get simpler tools, which limits environmental impact and wasted time.
Discernment: critical review
A person reviews anything they produce with AI assistance. We check facts and analysis against original sources and cite those, never AI itself. We train staff to spot bias in AI outputs. Our publications are all peer reviewed, and we ask peer reviewers not to rely on generative AI for their assessments.
Diligence: responsibility and transparency
Whoever produces a piece of work – whether a publication, a report or a social media post – remains fully responsible for it, whatever role AI played. Where funders or partners set stricter rules on AI, we follow them in the relevant work.
We are open about how we use AI. Standard uses need no separate acknowledgement. As outlined in our publication quality standards, these are language correction, copyediting and formatting; brainstorming and structuring; summarising source material; translation; and captions.
Where AI makes a material contribution to substantive work, such as analysis that shapes findings, the publication says so and explains how. AI-assisted translations are reviewed by a fluent speaker and labelled. AI-generated images are always labelled.
Looking ahead
We cannot yet anticipate every risk that comes with AI. We encourage experimentation and open sharing within safe limits. We support our staff through training as their roles and skills change. And of course, we will review our approach as technology, law and practice develop.
More broadly, AI is a societal shift that will likely change corruption and anti-corruption in ways that are hard to predict. Used responsibly and thoughtfully, it can help us spot those changes early, adapt quickly and develop better responses, in line with our research on anticipatory governance.
But technology alone will never solve corruption. At heart, it is a human problem – and it needs human judgement, grounded in strong values. That is why, however we use AI, our people and our values stay in charge.
Questions? Contact us at info@baselgovernance.org.