Scientists Asked AI to Impersonate 112 Public Figures. What Happened Next Is a ‘Dire’ Warning

2026-07-01 · Source: 404media Feed · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

A study published on Wednesday, July 1, 2026, in "PLOS One" revealed that AI chatbots, specifically GPT-4 Turbo, can generate responses perceived as more authentic, coherent, and relevant than actual public figures. Researchers prompted GPT-4 Turbo to impersonate 112 UK public figures, including politicians, business people, and journalists, using a dataset from BBC One's "Question Time" and Wikipedia biographies. A representative sample of 948 UK participants rated these AI-generated responses against verbatim human answers from the show. The findings, led by Steffen Herbold, a professor at the University of Passau, indicated that the LLM's content was consistently judged superior in authenticity, coherence, and relevance, even for well-known individuals. This outcome highlights a significant risk of public deception and accelerated misinformation in the political domain.

Key takeaway

For policy makers addressing election integrity and public trust, this study underscores the urgent need for regulatory action. Your constituents are highly susceptible to AI-generated political content, perceiving it as more authentic than real human statements. You should prioritize developing and implementing policies, such as banning political deepfakes. Additionally, fund public education campaigns to enhance media literacy regarding AI-generated messages. Failure to act risks accelerating misinformation and eroding democratic processes.

Key insights

AI-generated impersonations of public figures are perceived as more authentic and coherent than real human responses, posing a "dire" deception risk.

Principles

• LLMs can convincingly mimic public figures.
• Perceived authenticity is surprisingly easy to fake.
• AI coherence often surpasses human spontaneity.

Method

GPT-4 Turbo was prompted with a system message for mimicry and user prompts containing Wikipedia bios and "Question Time" questions. Responses were rated by 948 participants against real answers for authenticity, coherence, and relevance.

In practice

• Use LLMs for generating persuasive political content.
• Develop tools to detect AI-generated political speech.
• Inform public about AI's deceptive capabilities.

Topics

• AI Impersonation
• Political Deepfakes
• Misinformation Risk
• GPT-4 Turbo
• Public Deception
• Regulatory Policy

Best for: Research Scientist, CTO, VP of Engineering/Data, AI Scientist, AI Ethicist, Policy Maker

Related on AIssential

• We URGENTLY need a federal law forbidding AI from impersonating humans — Advanced deepfake and AI mimicry tools necessitate urgent federal laws to prevent "counterfeit people" and related scams.
• AI chatbots can effectively sway voters – in either direction — AI chatbots effectively sway political opinions by generating numerous factual claims, even if some are inaccurate.
• The risks of inviting AI into the heart of our economy, society and governance | Letters — The core risk of AI lies in its capacity to blur truth, erode provenance, and diminish human originality and empathy.
• "That's AI Slop, You Bot!" Studying Accusations, Evidence, and Credibility in Online Discourse Towards LLM-Generated Comments — Online "AI slop" accusations function as social gatekeeping for perceived authenticity, not accurate AI detection.
• ChatGPT and other AI bots made huge errors before Scottish election, study finds — AI chatbots frequently generate election misinformation, necessitating urgent regulatory intervention to protect democratic processes.
• Someone tried some prompt injection on me today 🤖 — The increasing prevalence of AI communication patterns blurs the lines of human authenticity in digital and real-world interactions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by 404media Feed.