The American intelligence community, along with think tanks and academics, are increasingly alarmed by the prospect of foreign entities gaining access to advanced AI capabilities.

Studies conducted by Gryphon Scientific and Rand Corporation have highlighted the potential dangers posed by large language models (LLMs) in particular. These models, drawing upon vast amounts of data, can provide detailed information that could be exploited by malicious actors, particularly in the realm of biological weapons development.

For instance, Gryphon’s research indicates that LLMs could offer sophisticated guidance at a post-doctoral level, assisting in troubleshooting issues related to the manipulation of dangerous viruses. Meanwhile, Rand Corporation’s findings suggest that LLMs could aid in the strategic planning and execution of biological attacks, such as by recommending efficient delivery methods for toxins like botulinum.

Furthermore, there’s growing concern that cyber actors could leverage AI to develop advanced tools for conducting large-scale, rapid, and elusive cyber attacks against critical infrastructure such as pipelines and railways.