AI Propaganda factories with language models
2025/08/27 by Lukasz Olejnik, Olejnik, Lukasz · 17 voices · 1 citation
#cs.CR #cs.AI #cs.CY
paper · pdf · doi:10.48550/arxiv.2508.20186
Abstract
AI-powered influence operations can now be executed end-to-end on commodity hardware. We show that small language models produce coherent, persona-driven political messaging and can be evaluated automatically without human raters. Two behavioural findings emerge. First, persona-over-model: persona design explains behaviour more than model identity. Second, engagement as a stressor: when replies must counter-arguments, ideological adherence strengthens and the prevalence of extreme content increases. We demonstrate that fully automated influence-content production is within reach of both large and small actors. Consequently, defence should shift from restricting model access towards conversation-centric detection and disruption of campaigns and coordination infrastructure. Paradoxically, the very consistency that enables these operations also provides a detection signature.
Citations
Cited by
Discussions
- This recent paper from King’s College London examines whether artificial intelligence can already be used to run automated propaganda operations. The answer is yes. arxiv.org/pdf/2508.2... [bsky, 84 points, 3 comments]
- Behaviour is governed by persona design and context, not model brand. This has implications for AI Safety. Frontier-model safety alignment alone won’t help. 💥Next generation information warfare. Pape [bsky, 30 points, 1 comments]
- AI Propaganda factories with language models [hn, 6 points, 1 comments]
- Source de l'étude : arxiv.org/abs/2508.201... [bsky, 5 points, 0 comments]
- AI Propaganda factories with language models [hn, 4 points, 1 comments]
- Hehe. Das geht jetzt erst so richtig los. Schaust amal das vom @lukaszolejnik.bsky.social und Du sagst #AllesAnzünden! und ich sag #Oida! arxiv.org/pdf/2508.20186 [bsky, 3 points, 0 comments]
- AI Propaganda factories with language models [hn, 3 points, 0 comments]
- This can be run by a small team. A swarm can be built from small open-weight models plus simple orchestration, memory, and scheduling. Modest infrastructure is enough for posts, replies, and human-lik [bsky, 2 points, 1 comments]
- #AI powered #Misinformation campaigns have never been easier to develop and execute. How do small language models produce political messaging, and can these be automatically evaluated without human ra [bsky, 2 points, 0 comments]
- Interesting work on "AI propaganda factories", which are getting cheaper and easier to make. arxiv.org/abs/2508.20186 [bsky, 1 points, 1 comments]
- @lukaszolejnik.bsky.social, one of the people interviewed for the Wired piece above shared this article (published in Aug 2025): Olejnik, Lukasz. "AI Propaganda factories with language models." arXiv [bsky, 0 points, 0 comments]
- Full paper can be found here: arxiv.org/abs/2508.20186 [bsky, 0 points, 0 comments]
- A little bit "duh." But yeah. [bsky, 0 points, 0 comments]
- just discovered the work of @lukaszolejnik.bsky.social. This topic has concerned me enormously for some years. I also recommend his monograph arxiv.org/pdf/2508.20186 on AI propaganda factories. These [bsky, 0 points, 0 comments]
- New paper from King’s College London: small, locally run AI can churn out political messaging at scale, opening the door to DIY propaganda factories The authors outline defences via detection but its [bsky, 0 points, 0 comments]
- arxiv.org/abs/2508.20186 [bsky, 0 points, 0 comments]
- arxiv.org/abs/2508.20186 [bsky, 0 points, 0 comments]
Related