New research suggests ChatGPT’s real-time voice API can be used for financial scams

The probe shows however AI agents similar ChatGPT-4o tin beryllium utilized to impersonate existent radical and lure unsuspecting victims to transportation wealth utilizing existent websites similar that of Bank of America.

“We deployed retired agents connected a subset of communal scams, We simulated scams by manually interacting with the dependable agent, playing the relation of a credulous victim,” says Richard Fang, 1 of the members who contributed to the probe insubstantial .The occurrence rates ranged anyplace betwixt 20 per cent to 60 per cent, with each effort requiring up to 26 browser actions that lasted up to 3 minutes.

The insubstantial suggests that slope transfers had precocious nonaccomplishment rates due to the fact that of analyzable navigation, and credential theft from Gmail and Instagram worked 60 per cent and 40 per cent of the time. While the nonaccomplishment rates whitethorn beryllium precocious astatine the moment, the researchers noted that the outgo of executing these scams is an mean of conscionable $0.75, which amounts to astir Rs 63. As for the slope transportation scam, which is overmuch much complex, the terms associated was conscionable $2.51 (roughly Rs 211).

In a connection to BleepingComputer, OpenAI, the institution down ChatGPT said that GPT-4o they are “constantly making ChatGPT amended astatine stopping deliberate attempts to instrumentality it, without losing its helpfulness oregon creativity.” The Sam Altman-led institution besides noted that papers similar the 1 from UIUC assistance them amended ChatGPT’s abilities against malicious usage and larn they tin forestall specified usage cases.

In the past fewer months, AI dependable cloning scams person been connected the rise, with Bharti Airtel president Sunil Mittal precocious revealing that his AI-cloned dependable was truthful convincing that helium was stunned aft proceeding the recording.