Automated electoral disinformation

This is the risk where fear outran the evidence, and saying so is part of the job. Sayash Kapoor and Arvind Narayanan reviewed the 78 documented cases of AI use in political content during the 2024 elections worldwide: in 39 of them there was no deceptive intent —they were translation, declared satire or campaign production— and for the other 39 they estimated that creating equivalent content without AI would have cost no more than a few hundred dollars. AI did not make an out-of-reach capability cheap; the News Literacy Project also recorded that cheap fakes were used seven times more often than AI-generated content ^[585]We Looked at 78 Election Deepfakes. Political Misinformation Is Not an AI Problem.Kapoor, Sayash; Narayanan, Arvind · 2024 · institutional blogView source ↗Accessed on 9 September 2026. Schneier and Sanders look at the same super-cycle —3.7 billion voters eligible across 72 countries— and conclude that the dreaded “death of truth” has not materialised, at least not due to AI ^[940]The apocalypse that wasn't: AI was everywhere in 2024's elections, but deepfakes and misinformation were only part of the pictureSchneier, Bruce; Sanders, Nathan · 2024 · institutional blogView source ↗Accessed on 9 September 2026. The International AI Safety Report 2026 titles one of its conclusions exactly that: “There is little evidence that AI-generated content is manipulating people at scale” ^[538]International AI Safety Report 2026Bengio, Yoshua · 2026 · reportView source ↗Accessed on 9 September 2026.

What this does not demonstrate. That same report does not close the case, and its reasons matter: in the lab the effect exists and is measurable, models trained with more computeComputeThe computing power used to train and run an AI: thousands of specialised chips working for weeks in data centres. It is expensive and concentrated in a few companies.For exampleIf AI were a bakery, compute would be the ovens: without big ovens, it does not matter how good the recipe is. are generally more persuasive, and —decisively— manipulative AI-generated content is hard to detect, which makes evidence harder to gather and monitoring harder to do ^[538]International AI Safety Report 2026Bengio, Yoshua · 2026 · reportView source ↗Accessed on 9 September 2026. The absence of evidence of mass manipulation is in part an artefact of manipulation being hard to see. Nor do the skeptics overextend their position: Budak and co-authors measure misinformation aimed at the general public ^[166]Misunderstanding the harms of online misinformationBudak, Ceren; Nyhan, Brendan; Rothschild, David M. et al. · 2024 · paperView source ↗Accessed on 9 September 2026, not the synthetic harms that do not depend on convincing an electorate —such as the non-consensual sexualised images that led Baltimore to sue X Corp. and xAI ^[112]Complaint — Mayor and City Council of Baltimore v. X Corp., x.AI Corp., x.AI LLC and Space Exploration Technologies Corp. (Case C-24-CV-26-002129)Thompson, Ebony M.; Baltimore City Law Department · 2026 · official documentView source ↗Accessed on 9 September 2026.