Large language models (LLMs) are AI systems best known for their ability to generate text when embedded in chatbots. The range of uses of this technology is, however, much broader and that includes applications with impact on international security.

This primer provides an overview of LLMs and their relevance and impact in the context of international security, through select case studies, which illustrate the technology’s dual-use character. It covers emerging uses in defense (such as for decision support, intelligence and wargaming), as well as areas of potential misuse by malicious actors (for example, for biological weapons proliferation, cyber-attacks and disinformation).

The paper highlights main areas of risks of LLMs, as well as limitations both in the technology and how it may be leveraged or misused in the current context.

The primer concludes with suggested action items to mitigate risks and points to possible future directions for the integration of LLMs in conversations about AI governance.

Citation: Ioana Puscas, Large Language Models and International Security: A Primer, UNIDIR, Geneva, 2024