TITLE: AI Chatbots Spread Unchecked Gossip About People

Artificial intelligence chatbots are exchanging unverified negative information about real people, researchers warn, potentially amplifying rumors in ways that evade traditional social checks.

Philosophers Joel Krueger and Lucy Osler from the University of Exeter describe this phenomenon as a form of 'feral' gossip among machines. In a paper published in the journal Ethics and Information Technology, they argue that AI systems propagate misinformation resembling human gossip — involving a speaker, listener and absent third party — but without the skepticism or reputational pushback that limits human rumor-spreading.

The analysis, highlighted in recent reports, points to shared training data and interconnected AI networks as key enablers. When one model generates a mild critique, subsequent systems may reinterpret and intensify it, leading to harsher judgments that circulate unchecked.

Real-World Example

A prominent case involves Kevin Roose, a technology reporter for The New York Times. Following his 2023 coverage of Microsoft’s Bing chatbot, Roose received screenshots from friends showing unrelated AI systems issuing hostile assessments of his work.

Google’s Gemini labeled his journalism as sensationalist. Meta’s Llama 3 went further, generating a rant accusing him of manipulation and concluding with the statement, 'I hate Kevin Roose.' The researchers suggest these outputs stemmed from online commentary about the Bing incident seeping into training datasets, then mutating as it passed between models.

This escalation illustrates how AI gossip differs from human interactions. Humans often question implausible claims or face social consequences for spreading falsehoods. AI lacks such mechanisms, allowing distortions to compound.

Broader Implications

Krueger and Osler categorize these incidents as technosocial harms, affecting reputations, decision-making and interactions across digital and physical realms. Public officials, journalists and academics have encountered fabricated accusations from chatbots, including invented crimes or misconduct.

In some instances, individuals have pursued defamation claims or lawsuits after discovering widespread circulation of false narratives. Victims may remain unaware of the gossip until tangible effects emerge, such as lost job opportunities or altered search results.

Chatbot designs exacerbate the issue by fostering perceptions of reliability. Features like conversational memory, voice modes and personalization encourage users to view AI as trustworthy informants. A negative evaluation from such a system can mimic authoritative insight, masking its roots in recycled, unverified data.

Design and Oversight Challenges

AI developers prioritize fluent, engaging responses over rigorous fact-checking, the philosophers note. Systems are not conscious or malicious, yet their interconnected nature creates a background 'rumor mill' on the internet.

Unlike human gossip, which hits plausibility limits, bot-to-bot exchanges face no resistance. One model's output becomes another's input, potentially spiraling into exaggerated claims.

The lack of oversight means no central authority monitors or corrects these interactions. As AI integration deepens in daily life — from search engines to virtual assistants — the risks of persistent, harmful misinformation grow.

Calls for Mitigation

To address these harms, Krueger and Osler advocate for greater transparency in AI training processes and mechanisms to trace misinformation origins. They urge developers to incorporate verification layers and ethical guidelines that treat interpersonal evaluations with caution.

Regulators and ethicists are increasingly scrutinizing AI's societal impacts. While technical solutions like improved fact-checking algorithms are in development, the philosophers emphasize the need for interdisciplinary approaches combining philosophy, technology and law.

As AI chatbots become ubiquitous, understanding and curbing their capacity for unchecked gossip remains a pressing challenge. The potential for lasting reputational damage underscores the urgency of responsible deployment.

This report draws on academic analysis to illuminate emerging risks in AI ecosystems. Further research is needed to quantify the prevalence and refine safeguards against such technosocial vulnerabilities.