When the target of a cyber operation is no longer the device — but the decision.
Artificial intelligence, behavioral analytics, synthetic media, automated agents and massive digital datasets are converging into a new information-security problem: the possibility of highly adaptive influence directed toward specific people, roles and decision environments.
REQUEST A CAMPAIGN →RESEARCH · AI PERSUASION
Peer-reviewed evidence, presented with its limitations. What the experiment demonstrated vs. what someone might infer.
Peer-reviewed research provides strong evidence that AI-generated text is persuasive and that personalization can increase persuasion. It does not demonstrate reliable control of an individual's decisions. The distinction between persuasion and control is central.
Finding: Personalized ChatGPT-generated messages were more influential than corresponding non-personalized versions across several tested contexts.
What it does not prove: Personalization can increase LLM persuasion in controlled settings.
Finding: GPT-4-generated messages were persuasive but no statistically significant overall microtargeting advantage from additional personal information was found.
What it does not prove: LLM persuasion is real; added microtargeting value is not clearly established.
Finding: Personalized GPT-4 debate conditions produced stronger persuasion than the human-human reference condition in that setting.
What it does not prove: Personalization can amplify persuasion relative to human persuaders under specific conditions.
Finding: LLM-generated political-policy messages caused measurable attitude changes and were approximately comparable to ordinary human-generated messages.
What it does not prove: LLM messages persuasive at parity with ordinary human messages, not necessarily superior.
Always separate what the experiment actually demonstrated from what someone might infer from it. A lab study showing personalized messages shift self-reported attitudes does not demonstrate that an autonomous system can reliably manipulate a specific official's consequential decision in the wild. The former is evidence; the latter is speculation.