So, it turns out the voice recognition system used by the Australian government can be fooled using artificial intelligence (AI). Am I the only one who feels like they’re about to be living the plot of Ex Machina?
An investigation by Guardian Australia found that the “voiceprint” feature used by both the Australian Tax Office (ATO) and Centrelink can be fooled by AI-generated voices.
The feature allows people to access their sensitive info over the phone after first verifying their voice. A Guardian Australia journo was able to trick the feature by using four minutes of audio to make a clone of their own voice. With this plus their customer reference number, they actually gainedaccess to their own self-service Centrelink account.
7.1 million Aussies have verified their voice with the ATO. Services Australia, the department that oversees Centrelink, saying on its website that voiceprint is “secure, accurate and reliable” – but with this new development I’m not convinced.
The website also states that: “It’s very difficult for someone to access your personal information. The system can tell when someone is pretending to be you or using a recording of your voice. We won’t give them access to your details.”
In some good news for the security of all your info, you do still need the customer reference number to access your accounts, which isn’t something that’s readily available to the public.
However, if someone did get their greasy little hands on your customer reference ID they could feasibly hack your account using voice-generated AI, which as a paranoid girlie, I don’t love.
The Guardian Australia contacted Services Australia with details of the potential security risk, but they declined to say if the voiceprint technology would be changed or removed from Centrelink.
A spokesperson, Hank Jongen, told Guardian Australia that Services Australia “has the capacity to continually assess risks and update processes accordingly” and that voice ID is a “highly secure authentication method” used by Centrelink.
“We continually scan for potential threats and make ongoing enhancements to ensure customer security,” he said.
Of course there are limitations to AI, like the fact that an AI-generated voice ~probably~ couldn’t respond to security questions quick enough to actually access any of your accounts.
AI and the use of it is a relatively new beast and it’s rapidly improving so who knows how good it could get.
An ATO spokesperson did assure Guardian Australia that they were regularly testing and improving security, including testing the risk of AI voice cloning.
While the voice generated is better with high-quality recordings, anyone with recordings of themselves speaking on social media, or who has had their voice recorded elsewhere, could be vulnerable to having their voice cloned by AI.
Now if you’ll excuse me, I’m going to crawl into a hole and cry about the AI takeover and then promptly turn off my cenno voiceprint.
