AI policies

[maelle]

Adding a ticket to track this here.

[maelle]

doi.org/10.5281/zenodo.17628864 shared by @rkrug

[maelle]

AI-powered coding and scripting assistants can support writing, debugging and optimising code used
for data processing, figure generation or text analysis. These tools can improve efficiency but require
careful validation. Experts must maintain full understanding of any code used in the assessment.
Coding and scripting support:
Usage: Supporting the writing and improving of code for data analysis, figure generation and
text extraction.
Risks: Inaccurate or insecure code; accidental disclosure of confidential data, access tokens
or API keys; reliance on opaque model logic; limited reproducibility.
Guidelines: Experts may use AI tools for code generation or debugging with caution.
Confidential data, credentials and access tokens must never be shared with AI systems. All
AI-generated code must be reviewed, tested and fully understood by the expert before use,
including the logic and statistical assumptions embedded in the code. Exceptions for local or
institutionally hosted tools may apply where confidentiality is ensured.
The use of AI-assisted coding must be documented in the DMR. Experts should also record
representative prompts or instructions used for code generation or debugging, including their
purpose, to support transparency and reproducibility. AI-generated code should not
independently determine or select statistical models; methodological decisions must remain
expert led.