“Generative Language Models and Automated Influence Operations: Emerging Threats and Potential Mitigations” was posted to arXiv in January 2023 by researchers from Georgetown University’s Center for Security and Emerging Technology, OpenAI, and the Stanford Internet Observatory, including Josh A. Goldstein, Girish Sastry, Micah Musser, Renee DiResta, Matthew Gentzel, and Katerina Sedova. It grew out of a 2021 workshop and assessed how large language models might reshape influence operations.
The report organizes the threat along three dimensions, the actors who run influence campaigns, their behaviors, and the content they produce, arguing that cheap, fluent text generation could let more actors run larger and more tailored operations. Its most-cited contribution is a “kill chain” framework that breaks the language-model-to-influence-operation pipeline into stages, from model construction and access through content generation and dissemination, and maps where mitigations could be applied at each stage. It deliberately avoids predicting a single outcome, instead laying out which interventions are promising, which are hard, and which carry trade-offs.
The paper was an early, structured attempt to think through generative-AI disinformation before the 2024 election cycle, and it framed much of the later policy and platform discussion about provenance, access controls, and detection.
For a business reader, it is a reference map of where AI-enabled disinformation can be disrupted, useful for anyone weighing how their tools, platforms, or content could be misused.