Propaganda for Hire: The Influence Industry in the Age of AI
There is an industry that will covertly run an influence campaign for anyone who pays — and AI has become its assembly line, mass-producing fake personas, multilingual posts, and even fake conversations. But here's what the panic misses: almost all of these campaigns flop. Inside the Dossier: how the influence-for-hire supply chain works, why AI boosts the propaganda but not its reach, and the one shift that could make today's "they all failed" record stop being true. Members only.
Key Judgement
We assess that a commercial "influence-for-hire" industry — private firms that run covert influence campaigns for paying clients — now operates alongside state propaganda services, and that generative AI has become its back-office productivity engine, cheaply manufacturing fake personas, multilingual content, images and even fake engagement at scale. Critically, and consistent with the threat reporting of the platforms that expose these operations, we assess that AI has so far boosted the supply of influence operations far more than their impact: the great majority fail to reach a meaningful audience, because distribution and authentic reach — not content production — remain the binding constraint.
We judge the principal near-term risks to be the commercialisation and deniability of the capability, and the erosion of the current detection chokepoint as operators migrate from monitored commercial AI services to self-hosted models — not a present-day collapse of the information environment.
Why This Dossier Matters
Warrier has mapped the factories behind consumer fraud and behind deepfakes. This maps the factory behind influence operations — the rentable firms and infrastructure that turn a client's agenda into a covert campaign. It also corrects a widespread misconception. The danger of AI in propaganda is usually imagined as hyper-persuasive fakes flooding and hypnotising the public. The documented reality is more mundane and more useful to understand: AI is largely a productivity tool for propagandists whose campaigns still mostly flop, and the real story is the industrialisation and commercialisation of the supply — together with the risk that today's comforting detectability will not last.
Source Notes
This Dossier draws on the influence-operation threat reporting of OpenAI and Meta. The commercial exemplar is STOIC, a Tel Aviv-based firm that OpenAI described as a "for-hire" threat actor and that both companies reported (May 2024) running a pro-Israel, anti-Hamas campaign targeting US and Canadian audiences using AI-generated personas, bios and images, and fake engagement; Meta removed 510 Facebook and 32 Instagram accounts and issued a cease-and-desist. The 2026 anchor is OpenAI's June 2026 report on PRC-linked operations ("Data Center Bandwagon" and "Tech and Tariffs") that used ChatGPT to inject narratives into US debates over AI data centres and tariffs, active late 2025–early 2026 with limited reach. Assessments that these operations did not meaningfully increase their audience — including Ben Nimmo's characterisation of the situation as "evolution, not revolution" and OpenAI's observation that AI changes the toolkit but not the error-prone operators — are from those reports, as is OpenAI's disclosure of disrupting 20-plus operations in 2024. State operations referenced (Russia's Doppelganger and Bad Grammar, China's Spamouflage, Iran's IUVM) are documented in the same reporting. These findings originate substantially with AI and platform companies, which have an interest in demonstrating that they police misuse.
MEMBERS ONLY — THE FULL DOSSIER FILE CONTINUES BELOW
🔒 This analysis is for warrier.ai Intelligence members only. → Become a Member
Already a member? Log in here
