Se l'IA inganna se stessa: i chatbot non riconoscono la grandissima parte dei video fatti con Sora
2026-01-23
La Repubblica.it
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
NewsGuard tested leading AI chatbots—ChatGPT, Gemini, and Grok—using videos generated by OpenAI's Sora. Without watermarks, the chatbots failed to identify AI-generated content in 78–95% of cases, sometimes confirming false events as real. This failure enables the spread of misinformation and undermines trust in AI detection systems.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves multiple AI systems (Sora for video generation and chatbots for detection) whose development and use have directly led to harm: the spread of misinformation through AI-generated videos that are not reliably detected by AI detectors. The failure of detection systems (malfunction) contributes to the harm by allowing false content to be accepted as real. This meets the definition of an AI Incident because it involves harm to communities through misinformation and a breach of trust in AI systems' reliability. The article documents actual harm (misinformation spread) and not just potential risk, so it is not merely a hazard or complementary information.[AI generated]