A voice identification system used by the Australian government, which allows people to verify their identity by phone, is vulnerable to attack by AI-generated voice software, according to The Guardian Australia. The service has been used by 3.8 million Centrelink clients and over 7.1 million people with the Australian Taxation Office (ATO) as of February 2023.
The technology creates a “digital representation of the sound, rhythm, physical characteristics and patterns of your voice”.
In addition to the voiceprint, callers must know an account holder’s customer reference number, which is included in communications such as childcare centre correspondence.
The vulnerability was highlighted by a journalist, who used four minutes of audio to clone their own voice, which was then used, alongside their customer reference number, to gain access to their own account.
The self-service phone system allows people to access sensitive information, including data on their benefits payments, and request documents to be sent by mail.
The voiceprint system is also used by the ATO, and at least one bank, Bank Australia, as an option during conversations with staff to reduce the need for verification questions. The automated system may be less vulnerable than other options to exploitation by AI-generated voice software, as the responses must be of high quality in real time. However, the technology is steadily improving.
Ed Santow, a former human rights commissioner and now director of policy at the Human Technology Institute at the University of Technology Sydney, commented that government agencies using biometrics as a form of verification needed to ensure they had the best systems in place, and that legislation should underpin those systems.
A spokesperson for the ATO stated that the agency had robust measures in place to protect its system from threats, including AI voice cloning.
A Nuance spokesperson, the company whose technology is used for the voiceprint service, referred to a February blog post in which it addressed the issue of “synthetic voices”.
