AI Advances Pose Biosecurity Opportunities, Risks: OpenAI

As AI models continue to evolve, their potential to accelerate scientific research in fields such as drug development, vaccine design, and biodefense grows considerably, reports OpenAI. However, AI models also present concerns over potential misuse, such as aiding in the recreation or development of biological threats.

AI systems have become increasingly integrated into biological research and development.   According to OpenAI, its most advanced models are approaching high capability levels in biology. These capabilities include reasoning over complex biological data, predicting chemical reactions, and designing experiments. For example, OpenAI and Babylon Biosciences have demonstrated that expert input can enhance the accuracy of large language models by collaborating to improve predictions about which early-stage drug candidates are most likely to succeed in clinical trials.

While these tools can assist in identifying drug candidates, developing diagnostics, and engineering enzymes for sustainable fuels, they can also be misused. As AI merges with emerging technologies in the life sciences, biosecurity risks are becoming increasingly complex and dynamic. To address this, OpenAI has implemented a multi-layered risk mitigation framework. The company has collaborated with experts in biosecurity, government entities, and national laboratories to shape its threat models, assess capabilities, and build safety protocols. These efforts include red teaming with domain specialists, creating always-on detection systems, enforcing usage policies, and applying strict access controls to model data and weights.

OpenAI emphasizes the importance of proactive measures. As part of its Preparedness Framework, the company has trained its models to withhold step-by-step instructions or troubleshooting guidance for dual-use biological requests. This approach aims to ensure that while experts can access useful insights, misuse by individuals without appropriate knowledge is minimized.

Monitoring systems have also been scaled across platforms using these models. Responses flagged as unsafe are blocked automatically and subject to review. Repeated or egregious misuse can lead to account suspensions or, in certain cases, notifications to law enforcement.

In July, OpenAI will host a biodefense summit with government researchers and NGOs to examine the dual-use risks and collaborative opportunities presented by frontier AI models. The summit aims to explore how AI can support the development of diagnostics, therapies, and countermeasures in a secure manner.

To address challenges arising from unregulated access to powerful AI tools, OpenAI calls for stronger coordination between the public and private sectors. Potential actions include improving pathogen detection systems, enhancing nucleic acid synthesis screening, and investing in next-generation biosecurity infrastructure.