Can AI fool people into believing fabricated or hallucinated information ?
Esprimi il tuo voto — poi leggi cosa hanno trovato la nostra redazione e i modelli di IA.
Can artificial intelligence be used to deceive by fabricating or hallucinating convincing yet false information? Generative AI systems now produce text indistinguishable from human writing, a capability increasingly exploited in phishing, propaganda, and synthetic media.
Background
AI can produce convincing fabricated or hallucinated information that people often accept as true, especially when the output is tailored to sound authoritative or emotionally resonant. Systems like large language models can generate text indistinguishable from human writing, which has been exploited in phishing, misinformation campaigns, and deepfake content creation. Studies show that humans are prone to trusting AI-generated text despite its inaccuracies, particularly when it aligns with their existing beliefs or is presented with confidence. However, AI lacks true understanding, relying on patterns in training data rather than factual verification, which can lead to plausible but false assertions. Detection tools exist but are not foolproof, as adversaries continuously refine methods to bypass safeguards. Social media platforms and regulators have struggled to keep pace with the spread of AI-generated disinformation, which can erode public trust and influence real-world behavior. The risk is amplified when AI systems are fine-tuned on biased or low-quality data, further distorting outputs.
Suggerisci un tag
Manca un concetto su questo tema? Suggeriscilo e un amministratore lo valuterà.
Stato verificato l'ultima volta il May 15, 2026.
Galleria
Can AI fool people into believing fabricated or hallucinated information?
La giuria ha trovato una risposta chiaramente affermativa.
Avendo valutato le prove, la giuria ha stabilito che i moderni modelli linguistici avanzati possono effettivamente creare falsificazioni persuasive e simili a quelle umane, abbastanza convincenti da ingannare, anche se non arrivano a una perfetta imitazione. L’unico quasi-voto ha avvertito che la capacità di convincere non sempre equivale al successo, soprattutto quando il pubblico si ferma a porre le domande giuste. Il tribunale conclude che il pericolo è reale e la difesa non regge.
Having weighed the evidence, the jury found that today’s advanced language models can indeed craft persuasive, human-sounding fabrications convincing enough to mislead, though they could stop short of a perfect impersonation. The lone almost-vote warned that convincingness does not always equal success, especially when audiences pause to ask the right questions. The bench concludes that the danger is real, and the defense is not.
But the data is real.
The Case File
By a vote of 3 — 1 — 0, the panel returns a verdict of Sì, with verdict confidence of 85%. The court so orders.
"Advanced language models can generate convincing text"
"AI can generate highly convincing but unverified content, often indistinguishable from human-created misinformation."
"Advanced LLMs can generate persuasive, coherent hallucinations that deceive users in controlled settings."
"Advanced language models can generate convincing text 2020-06"
Le singole dichiarazioni dei giurati sono mostrate nell'inglese originale per preservare la precisione probatoria.
Cosa pensa il pubblico
No 0% · Sì 67% · Forse 33% 3 votesDiscussione
no comments⚖ 1 jury check · più recente 7 ore fa
Ogni riga è un controllo di giuria separato. I giurati sono modelli di IA (identità tenute volutamente neutre). Lo stato riflette il conteggio cumulativo su tutti i controlli — come funziona la giuria.