Since July, researchers at Los Alamos National Laboratory in New Mexico have been exploring how the AI model GPT-4o, developed by OpenAI, can assist in biological research tasks. These tasks include maintaining cell cultures, separating cells using a centrifuge, and introducing genetic material into organisms. The goal of these evaluations is to innovate biosciences and understand potential risks, particularly focusing on biosafety and biosecurity issues since AI models like ChatGPT became publicly available in 2022. At Johns Hopkins Center for Health Security, experts emphasize the potential of AI in improving public health but also note the technology's unpredictable risks. They urge governments to identify and mitigate the most pressing risks, leveraging past experience in managing biological research risks. GPT-4o is a multimodal large language model (LLM) capable of processing text, audio, image, and video inputs. Previously, GPT-4 demonstrated applications in life sciences, such as designing antibodies and automating biology experiments.

However, these advances could also pose significant biosafety risks, pushing governments to establish AI safety institutes and implement guidelines to manage such risks. To determine which AI capabilities might lead to large-scale harm, experts suggest focusing on those that enable pandemic-level outbreaks, such as optimizing virus designs or automating pathogen synthesis. They call for broad cooperation between AI developers and safety experts to prioritize mitigating severe risks. Safety evaluations of AI models often focus narrowly on the development of bioweapons without differentiating between smaller and larger-scale risks. Evaluations also need to account for the collaborative operation of multiple AI systems, rather than concentrating solely on individual models. A more effective approach requires an objective scientific consensus on which AI capabilities pose significant biosafety threats. Building such consensus would involve thorough safety testing, developed by independent experts, and would be supported by both government and industry stakeholders. This approach ensures a balance between harnessing AI's benefits and mitigating its risks.