Responsible AI

How Aeviant uses AI in research, checks AI-assisted work and limits the information submitted to external services.

These rules apply when Aeviant uses AI to search or organize scientific literature, write or inspect code, examine computational work or prepare scientific material.

Uses of AI

Aeviant uses AI to search and organize scientific literature, write or review code, examine computational methods and outputs, and draft or edit research text.

Aeviant authorizes individuals to use research AI systems only for specified research or operational work. Access is individually assigned under Aeviant’s Security and controlled access statement.

An AI system may propose a citation, explanation, code change or interpretation. Aeviant does not rely on the output for a public scientific claim or program decision until it completes the relevant check described below.

Checking AI-assisted work

Aeviant checks citations against the cited source and scientific statements against primary literature. For AI-assisted code, Aeviant reviews the intended method and inputs, runs the code and inspects the output. For AI-assisted calculations, Aeviant checks the result against the underlying data and method.

If an output cannot be checked, Aeviant does not use it as the basis for a public scientific claim or program decision.

Records and public disclosure

For AI-assisted analysis or code used in a program decision, Aeviant records the system, task, output used and checks performed.

Aeviant does not present AI-generated data, images or results as observations produced by an experiment. When AI-generated analysis, data or imagery appears in public scientific material, Aeviant identifies it as AI-generated.

Approval and concerns

Aeviant assigns responsibility for approving research AI services and responding when they are used outside their authorized work or access restrictions.

Concerns about inappropriate AI use or related safety or security issues may be sent to [email protected]. Aeviant records and reviews the report and may suspend affected access while the review is underway.

Confidential research

Non-public research is submitted only through an AI service and account configuration approved for that information under Aeviant’s external-service review.

Before approval, Aeviant checks who can access submitted information, how long the provider retains it, whether the provider may use it to train or improve models and which account and sharing controls apply.

Aeviant does not submit the research if the service’s terms or account controls conflict with the restrictions on that research.

The treatment of unpublished and patent-sensitive research is described in Aeviant’s Security and controlled access statement.

Last reviewed .