OpenAI announced a research partnership with Los Alamos National Laboratory on 10 July 2024, focused on evaluating AI models for bioscience research applications [source]. The collaboration will assess how OpenAI's models perform in scientific workflows, with Los Alamos scientists testing the technology for tasks including literature review, data analysis, and hypothesis generation.

The partnership includes a safety component. Los Alamos will provide feedback on potential biosecurity risks associated with AI-assisted research, informing OpenAI's development of future safeguards. The laboratory's expertise in national security research positions it to identify dual-use concerns that may emerge from increasingly capable models.

OpenAI stated the work will help establish best practices for deploying AI in sensitive research environments. Los Alamos scientists will document cases where models produce inaccurate scientific information or suggest unsafe experimental procedures. This feedback loop aims to improve model reliability before broader deployment in laboratory settings.

The announcement follows growing scrutiny of AI applications in biological research. Policymakers and researchers have raised concerns about models trained on scientific literature potentially synthesising information in ways that lower barriers to harmful applications. OpenAI has previously conducted internal biosecurity evaluations but has not published detailed methodologies or results.

No technical specifications were provided about which OpenAI models will be evaluated or what benchmarks will measure performance. The partnership timeline and publication plans for findings were not disclosed. Los Alamos confirmed the collaboration in a separate statement but did not detail which research groups would participate or what resources the laboratory would commit to the evaluation process.