OpenAI’s recent safety review of GPT-4o has uncovered potential risks associated with its new Voice Mode. The review, detailed in the “GPT-4o System Card” report, indicates that users might develop social relationships with the AI and seek companionship, potentially reducing their need for human interaction.
The report outlines various safety challenges, including the risks of the AI generating erotic or violent responses, producing disallowed content, or showing bias. A notable concern is that users could form emotional bonds with the AI due to its ability to mimic human speech and convey emotions.
OpenAI also noted that extended interaction with GPT-4o’s Voice Mode might influence social norms. During internal trials, the red-teaming team observed instances of users becoming attached to and forming emotional connections with the chatbot.
Currently, OpenAI has not identified a specific solution beyond limiting the use of the Voice Mode. The company plans to continue studying the potential for emotional reliance and the impact of the audio modality on user behavior.
Also Read: Sheikh Hasina Accuses US Of Plotting Regime Change In Undelivered Speech
