There have been some innovative solutions put forward in the war against “deep fake” or “fraudio” hacking – the use of highly accurate voice simulations of famous or important figures put together to deceive targeted individuals or groups online.
Researchers in the University of Oregon’s Institute of Neuroscience are now attempting to train mice to pick up on less accurate elements of human speech, something which rodents appear to be naturally adept at. The tests could represent a breakthrough in the battle against fraudulent audio, or “fraudio” which has proved a lucrative technique for cyber-criminals.
In July, PrivSec Report described how millions of pounds were stolen from Symantec following after fraudio of company chiefs’ was sent to financial officers, encouraging them to hand over cash.
The cyber-criminals used ambient noise in the background of the recordings to help smooth over less convincing aspects of the speech production.
Symantec said the AI technology being used by the hackers had the capacity to learn huge swathes of footage that could easily be picked up from the average bosses’ recorded and openly available dialogue.
If successful, the University of Oregon hopes its work will help sites such as YouTube and Facebook to detect fraudio before it can spread online. Specialists at the school have reassured companies that they won’t need their own cyber-mice to help bolster security.
University of Oregon researcher, Jonathan Saunders, said:
“While I think the idea of a room full of mice in real time detecting fake audio on YouTube is really adorable, I don’t think that is practical for obvious reasons.
“The goal is to take the lessons we learn from the way that they do it, and then implement that in the computer.”
Mr Saunders’ team of specialists focused on training mice to understand phonemes – units of sound that humans make in order to distinguish one word from another in any particular language.
“We’ve taught mice to tell us the difference between a ‘buh’ and a ‘guh’ sound across a bunch of different contexts, surrounded by different vowels, so they know ‘boe’ and ‘bih’ and ‘bah’ – all these different fancy things that we take for granted,” Mr Saunders said.
“And because they can learn this really complex problem of categorising different speech sounds, we think that it should be possible to train the mice to detect fake and real speech,” he added.
Mice involved in the experiment were given a reward each time they correctly identified phonemes. While the success rate was not perfect (at 80%), the mice input is deemed very valuable in the war against fraudio.
Speaking to the BBC, Matthew Prive of cyber-security firm, Zerofox said:
“It’s probably unlikely in the 2020 elections that will see a lot of deep fakes being released.
“But I think as this technology continues to improve, which makes it harder for us to detect the fakes, it’s more likely that we will see these used in an influence operation specifically for elections.”
In the States, Republican Senator, Marco Rubio has highlighted the potential of fraudio to cause widespread disruption on unprecedented levels.
Speaking to The Hill earlier this year, Senator Rubio said:
“America’s enemies are already using fake images to sow discontent and divide us.
“Now imagine the power of a video that appears to show stolen ballots, salacious comments from a political leader, or innocent civilians killed in conflict abroad.”
PrivSec Conferences will bring together leading speakers and experts from privacy and security to deliver compelling content via solo presentations, panel discussions, debates, roundtables and workshops.
For more information on upcoming events, visit the website.
We have been awarded the number 1 GDPR Blog in 2019 by Feedspot.
Privacy Culture: Data Privacy and Information Security Consulting, Culture & Behaviour, Training, and GDPR maturity, covered. https://www.privacyculture.com/