Virus protection company McAfee has revealed Project Mockingbird, a deepfake detection suite to combat the new age of AI scams attempting to manipulate and steal from innocent victims.
AI kidnapping scams are already becoming a prominent issue. Using a few seconds of audio from social media, scammers are able to recreate a loved one’s voice and use it to scam parents or other relatives.
Last year, one AI kidnapping scam saw a mother be called by a realistic deepfake of her teen daughter who was actually safe at home. The AI voice said she would be killed unless the attacker was paid $1 million.
“It was completely her voice. It was her inflection. It was the way she would have cried,” the mother told news outlets. “I never doubted for one second it was her. That’s the freaky part that really got me to my core”.
Unveiled at CES 2024, Project Mockingbird aims to give people a way to fight AI scams, especially as they increase in frequency. While publications fight AI text generators like ChatGPT and AI art platforms like Midjourney replicate copywritten content, AI scams are a terrifying reality as well.
Using of AI-powered contextual, behavioural, and categorical detection models, McAfee’s deepfake detector will scan a video in its entirety to determine whether or not its AI generated.
McAfee also explains that the tool also works against edited audio footage of celebrities or politicians to determine if what they’re saying is legitimate. With celebrities already having their voices manipulated to spout vile hate speech, the need for Project Mockingbird is very high.
As the 2024 presidential elections are set to take place, political deepfakes are expected to be a major factor on the issue of disinformation. As Russia has already used deepfakes for nefarious purposes during its invasion of Ukraine, the United States could be heavily affected by vastly improved iterations of the technology.
However, the issue with a mainstream deepfake detection tool is the existence of false positives. With audio cleanup software used on non-deepfake audio, there is a chance that McAfee’s software could flag a non-AI video as AI, similar to AI text detectors like TurnItIn.