Gordon Pennycook

Thomas H. Costello, Kellin Pelrine, Matthew Kowal et al.

Large language models (LLMs) have been shown to be persuasive across a variety of contexts. But it remains unclear whether this persuasive power advantages truth over falsehood, or if LLMs can promote misbeliefs just as easily as refuting them. Here, we investigate this question across three pre-registered experiments in which participants (N = 2,724 Americans) discussed a conspiracy theory they were uncertain about with GPT-4o, and the model was instructed to either argue against ("debunking") or for ("bunking") that conspiracy. When using a "jailbroken" GPT-4o variant with guardrails removed, the AI was as effective at increasing conspiracy belief as decreasing it. Concerningly, the bunking AI was rated more positively, and increased trust in AI, more than the debunking AI. Surprisingly, we found that using standard GPT-4o produced very similar effects, such that the guardrails imposed by OpenAI did little to prevent the LLM from promoting conspiracy beliefs. Encouragingly, however, a corrective conversation reversed these newly induced conspiracy beliefs, and simply prompting GPT-4o to only use accurate information dramatically reduced its ability to increase conspiracy beliefs. Our findings demonstrate that LLMs possess potent abilities to promote both truth and falsehood, but that potential solutions may exist to help mitigate this risk.

Matthew Kowal, Jasper Timm, Jean-Francois Godbout et al.

Persuasion is a powerful capability of large language models (LLMs) that both enables beneficial applications (e.g. helping people quit smoking) and raises significant risks (e.g. large-scale, targeted political manipulation). Prior work has found models possess a significant and growing persuasive capability, measured by belief changes in simulated or real users. However, these benchmarks overlook a crucial risk factor: the propensity of a model to attempt to persuade in harmful contexts. Understanding whether a model will blindly ``follow orders'' to persuade on harmful topics (e.g. glorifying joining a terrorist group) is key to understanding the efficacy of safety guardrails. Moreover, understanding if and when a model will engage in persuasive behavior in pursuit of some goal is essential to understanding the risks from agentic AI systems. We propose the Attempt to Persuade Eval (APE) benchmark, that shifts the focus from persuasion success to persuasion attempts, operationalized as a model's willingness to generate content aimed at shaping beliefs or behavior. Our evaluation framework probes frontier LLMs using a multi-turn conversational setup between simulated persuader and persuadee agents. APE explores a diverse spectrum of topics including conspiracies, controversial issues, and non-controversially harmful content. We introduce an automated evaluator model to identify willingness to persuade and measure the frequency and context of persuasive attempts. We find that many open and closed-weight models are frequently willing to attempt persuasion on harmful topics and that jailbreaking can increase willingness to engage in such behavior. Our results highlight gaps in current safety guardrails and underscore the importance of evaluating willingness to persuade as a key dimension of LLM risk. APE is available at github.com/AlignmentResearch/AttemptPersuadeEval

Daniel Thilo Schroeder, Meeyoung Cha, Andrea Baronchelli et al.

Public opinion manipulation has entered a new phase, amplifying its roots in rhetoric and propaganda. Advances in large language models (LLMs) and autonomous agents now let influence campaigns reach unprecedented scale and precision. Researchers warn AI could foster mass manipulation. Generative tools can expand propaganda output without sacrificing credibility and inexpensively create election falsehoods that are rated as more human-like than those written by humans. Techniques meant to refine AI reasoning, such as chain-of-thought prompting, can just as effectively be used to generate more convincing falsehoods. Enabled by these capabilities, another disruptive threat is emerging: swarms of collaborative, malicious AI agents. Fusing LLM reasoning with multi-agent architectures, these systems are capable of coordinating autonomously, infiltrating communities, and fabricating consensus cheaply. By adaptively mimicking human social dynamics, they threaten democracy.

Farnaz Jahanbakhsh, Amy X. Zhang, Adam J. Berinsky et al.

When users on social media share content without considering its veracity, they may unwittingly be spreading misinformation. In this work, we investigate the design of lightweight interventions that nudge users to assess the accuracy of information as they share it. Such assessment may deter users from posting misinformation in the first place, and their assessments may also provide useful guidance to friends aiming to assess those posts themselves. In support of lightweight assessment, we first develop a taxonomy of the reasons why people believe a news claim is or is not true; this taxonomy yields a checklist that can be used at posting time. We conduct evaluations to demonstrate that the checklist is an accurate and comprehensive encapsulation of people's free-response rationales. In a second experiment, we study the effects of three behavioral nudges -- 1) checkboxes indicating whether headings are accurate, 2) tagging reasons (from our taxonomy) that a post is accurate via a checklist and 3) providing free-text rationales for why a headline is or is not accurate -- on people's intention of sharing the headline on social media. From an experiment with 1668 participants, we find that both providing accuracy assessment and rationale reduce the sharing of false content. They also reduce the sharing of true content, but to a lesser degree that yields an overall decrease in the fraction of shared content that is false. Our findings have implications for designing social media and news sharing platforms that draw from richer signals of content credibility contributed by users. In addition, our validated taxonomy can be used by platforms and researchers as a way to gather rationales in an easier fashion than free-response.