As the Secretariat for ISDC and SPIA, and implementing CGIAR’s independent evaluation function, IAES has adopted a set of principles providing a common framework for responsible use of AI. The idea is to transition the use of AI across ISDC, SPIA and IAES from cautious coexistence toward confident, transparent, and responsible use – as noted in our introductory blog earlier this year.
Setting our course: a roadmap to support experimentation and reflection
We have been actively exploring how emerging technologies can support our work under an internal roadmap. In 2025, the IAES/Evaluation Function issued its Technical Note on Use of AI in Evaluations, pioneering the approach among its peers. ISDC has made the testing of Large Language Models (LLM) and their role in advisory processes a key deliverable of the 2026 workplan. SPIA’s use of evidence Senior Officer has led a cross-functional experiment earlier this year on testing AI production of donor-tailored summaries of studies, evaluations, and independent science advice, alongside the Panel’s ongoing work to develop its approach to AI.
Many pathways converge: common principles
Because these activities share many features, and implementation to date has yielded similar lessons, establishing a common denominator made perfect sense. With IAES operating within CGIAR and hosted in the European Union by the Alliance of Bioversity International and CIAT, we could draw on existing policies and legislation to develop principles broad enough to cover all three workstreams, yet focused enough to guide our use of AI.
The seven principles – aligning for a clear direction ahead
The approach is grounded in human agency and oversight, ensuring that AI complements rather than replaces human expertise and judgment, and transparency, promoting appropriate traceability, explainability and disclosure of contributions. Technical capacity and fitness for purpose emphasize the careful selection and testing of reliable tools suited to specific tasks and staff training, with privacy and data governance safeguarding personal and sensitive information in line with applicable requirements.
At the same time, fairness and inclusion call for identifying and mitigating potential bias and preserving diverse perspectives, knowledge and contexts. All the while, social and environmental responsibility encourage consideration of AI’s wider costs alongside the efficiencies and value it can bring. Underpinning all of these principles are independence and accountability, ensuring that the use of AI does not compromise the quality, and that responsibility for final outputs remains firmly with people and the relevant institutional leadership.
From principles to practice
With the principles now adopted, the Advisory Bodies have a common reference point. Protocols and standard clauses will be further developed and integrated into implementation of the 2027–2028 workplan, but their real value will come from how they guide our day-to-day work. As a starting point, we drafted some examples of tasks that may be supported by AI, their indicative risk levels, and the corresponding expectations for human review and disclosure, to embed in our IAES risk matrix. As AI use evolves, IAES will continue to build AI literacy and document emerging lessons and concerns, preserving the quality, credibility or independence of scientific advisory and evaluative work at all times.
