Reflection and Transfyr to test open AI models in science

Reflection and Transfyr say they are partnering to evaluate open AI models in scientific settings, examining how AI assistance affects research and whether safeguards hold up as experiments unfold. The announcement describes planned evaluations, not completed tests or results. ^[1]

Reflection says the work will use execution data—records of actions and experimental outcomes—to assess where model guidance helps, how it affects results, and whether safety boundaries hold during interactions. The companies say they will consider whether and how to share methods and findings responsibly; the announcement does not specify a timeline, models, or experimental settings. ^[1]