What happened

Researchers at the University of Illinois Urbana-Champaign demonstrated that OpenAI's realtime voice API can be exploited to automate phone scams at a remarkably low cost - around USD 0.75 per successful scam. 

The API allows developers to create AI agents capable of impersonating various figures, such as bank officials or government agents, to extract sensitive information from victims. 

The study revealed that these AI agents could autonomously execute a range of scams, including credential theft and bank account hijacking, with varying success rates.

Why it happened

Recent advancements in OpenAI's technology, specifically the release of its Realtime API, which enables real-time interaction through voice, means anyone can mimic and deloy someone else's voice for any purpose.

Concerns have been raised about the potential for misuse when OpenAI demonstrated voice mimicry without consent, leading to a delay in the release of its advanced voice features. 

Despite OpenAI's implementation of some safety measures, researchers found that the API's design was simple enough to allow for easy exploitation by malicious actors. 

The study highlighted that phone scams already affect millions of Americans annually, costing around USD 40 billion, making this new tool particularly concerning.