The company assessed the persuasiveness of GPT-4o’s text and voice features. Launched publicly in May, GPT-4o's voice modality was deemed low risk, while the text modality slightly exceeded the threshold for medium risk, according to their research paper.
