NewsGuard offers ‘misinformation risk audit’ to generative AI companies
In brief
News reliability ratings company NewsGuard is offering a new service using trained analysts to audit generative AI companies.
The audit will see NewsGuard analysts test chatbots’ performance in promoting or rejecting misinformation, with prompts drawn from a proprietary catalogue of more than 1,200 false narratives.
Publishers and watchdogs have expressed broad anxieties about various chatbots’ ability to produce fake or misleading information at pace and scale. NewsGuard’s own analysis of ChatGPT-3.5, the chatbot model before the updated GPT4, found that the AI complied with a request to propagate a false narrative four out of five times.
