In recent years, analysts and scholars have noted that corruption or dysfunction in the global information ecosystem could have the effect of increasing nuclear risk.[1] In these works, corruption and dysfunction are interpreted broadly to include mis- and dis- information but also other information-related phenomena such as provocative or intemperate content, true-but-misleading information, or attentional diversions.

Now, the recent explosion in the capabilities of artificial intelligence—specifically, large language models (LLMs)—has led to the automated generation of novel text in enormous volumes and, increasingly, images and videos. For those interested in the creation of content to corrupt the information environment, there may be no better tool than a large language model. LLMs put former approaches to generating such content to shame. Also, LLM tools require far less expertise to produce and use, making the capabilities they afford much more broadly accessible and therefore expanding the number of possible threat actors as promoters of information dysfunction.

Continue reading at thebulletin.org.