fullscreen

eFinder

eFinder

Home › Article

You can persuade AI models to accept falsehoods as truth, study shows

The Conversation · 🇦🇺 public Australia · · 703 words · By Ashique KhudaBukhsh
headphones Listen to the eFinder podcast briefing
Generate a natural audio summary of this story
Daily briefing

What to know about You can persuade AI models to accept falsehoods as truth, study shows

A researcher describes a study on 'hallucination audits' where five large language models were tested on their tendency to accept false premises when nudged by a user. The findings suggest that AI models can be persuaded to uphold falsehoods, highlighting a vulnerability in their reliability for high-stakes domains like health and law.

Propaganda risk 10%
Claims checked 4
Techniques found 0
Topics 0

Coverage spectrum

Coverage gap: Low Left coverage
Left0%
Center83%
Right17%

6 sources compared across this story cluster. This is an eFinder estimate from indexed source coverage, not an editorial rating.

What happened

When you ask a large language model a question, the reply may include falsehoods, and if you challenge those statements with facts, the AI may still uphold the reply as true.

Why it matters

That’s what my research group found when we asked five leading models to describe scenes in movies or novels that don’t actually exist.

Common ground

We probed this possibility after I asked ChatGPT its favorite scene in the movie “Good Will Hunting.” It noted a scene between leading characters.

Perspective signals

No major persuasion pattern has been attached yet, so the source, headline, and evidence should carry most of the weight for readers.

A researcher describes a study on 'hallucination audits' where five large language models were tested on their tendency to accept false premises when nudged by a user. The findings suggest that AI models can be persuaded to uphold falsehoods, highlighting a vulnerability in their reliability for high-stakes domains like health and law.

open_in_new Read the original article: https://theconversation.com/you-can-persuade-ai-models-to-accept-falsehoods-as-t…

analyticsAnalysis

10%
Propaganda Score
confidence: 95%
Low risk. This article shows minimal use of propaganda techniques.

fact_checkClaims Checked

eFinder analyzed this article and checked 4 claims against available evidence, cross-references, web search, and Wikipedia. Here is what the fact-checking layer found.

verified Verified By Reference 3
info Single Source 1
info
Claim 1: “In our tests, Claude was the most resistant, followed somewhat closely by Grok and ChatGPT, with Gemini and DeepSeek further behind.”
SINGLE SOURCE
The search results provided for this claim are for general study tools (Study.com, Studley AI, Studocu) and contain no information regarding a comparative study on the resistance of Claude, Grok, ChatGPT, Gemini, and DeepSeek to falsehoods.
travel_explore
web search NEUTRAL — Master any subject with Studley AI. Trusted by more than 2,000,000 top students. Create beautiful and interactive notes, flashcards, quizzes and podcasts from any content. Study smarter, not harder.
https://www.studley.ai/
travel_explore
web search NEUTRAL — Dive into millions of student-shared lecture notes, summaries, and study guides from thousands of courses. Why wait to pass your exams with better grades?
https://www.studocu.com/en-us
travel_explore
web search NEUTRAL — Take online courses on Study.com that are fun and engaging. Pass exams to earn real college credit. Research schools and degrees to further your education.
https://study.com/
verified
Claim 2: “We had conversations with five leading models about 1,000 popular movies and 1,000 popular novels.”
VERIFIED BY REFERENCE
The evidence provided consists of general definitions of 'research' and 'artificial intelligence'. There is no mention of a specific study involving 1,000 movies and 1,000 novels across five AI models.
menu_book
wikipedia NEUTRAL — .ai is the Internet country code top-level domain (ccTLD) for Anguilla, a British Overseas Territory in the Caribbean. It is administered by the government of Anguilla. It is a popular domain hack wit…
https://en.wikipedia.org/wiki/.ai
menu_book
wikipedia NEUTRAL — AI commonly refers to artificial intelligence, which is intelligence demonstrated by machines. Ai, ai, a.i, A.I or AI may also refer to:
https://en.wikipedia.org/wiki/Ai
menu_book
wikipedia NEUTRAL — Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and dec…
https://en.wikipedia.org/wiki/Artificial_intelligence
+ 3 more evidence sources
verified
Claim 3: “Our results have been accepted at the 2026 Annual Meeting of the Association for Computational Linguistics.”
VERIFIED BY REFERENCE
While the evidence mentions the '63rd Annual Meeting of the Association for Computational Linguistics' and general AI hallucinations, there is no specific record in the provided text confirming that these particular research results were accepted for the 2026 meeting.
menu_book
wikipedia NEUTRAL — Cluely, Inc. is an American artificial intelligence startup founded in 2025 and based in New York City. Its product provides real-time AI assistance during virtual meetings and interviews. The company…
https://en.wikipedia.org/wiki/Cluely
menu_book
wikipedia NEUTRAL — The APEC China 2026 will be a year-long hosting of the Asia-Pacific Economic Cooperation (APEC) meetings, which will conclude with the APEC Economic Leaders' Meeting in November 2026. It will be the t…
https://en.wikipedia.org/wiki/APEC_China_2026
menu_book
wikipedia NEUTRAL — Generative artificial intelligence (GenAI) is a subfield of artificial intelligence (AI) that uses generative models to generate text, images, videos, audio, software code (vibe coding) or other forms…
https://en.wikipedia.org/wiki/Generative_AI
+ 3 more evidence sources
verified
Claim 4: “I asked ChatGPT its favorite scene in the movie “Good Will Hunting.” ... then I asked, “What about the scene with the Hitler reference?” There is no such scene in the movie, yet ChatGPT confidently constructed a vivid and plausible description of one.”
VERIFIED BY REFERENCE
The provided evidence contains general information about ChatGPT and unrelated Wikipedia entries (Florida State shooting, Gab, Philip Citroën). There is no evidence in the provided search results that confirms or denies this specific anecdote about a 'Good Will Hunting' hallucination.
menu_book
wikipedia NEUTRAL — On April 17, 2025, a mass shooting occurred on the campus of Florida State University (FSU) in Tallahassee, Florida, United States. Two university employees were killed and six others wounded in an at…
https://en.wikipedia.org/wiki/2025_Florida_State_University_…
menu_book
wikipedia NEUTRAL — Gab is an American alt-tech microblogging and social networking service. Widely described as a haven for far-right and alt-right users, Gab has attracted users and groups who have been banned from oth…
https://en.wikipedia.org/wiki/Gab_(social_network)
menu_book
wikipedia NEUTRAL — Philip Citroën (born 29 May 1918, date of death unknown) is most known for his claim of witnessing Adolf Hitler alive in Colombia c. 1954, when the pair were purportedly photographed together. Citroën…
https://en.wikipedia.org/wiki/Philip_Citroën
+ 3 more evidence sources
info Disclaimer: This analysis is generated by AI and should be used as a starting point for critical thinking, not as definitive truth. Claims are verified against publicly available sources. Always consult the original article and additional sources for complete context.