At CES 2024 this week, cybersecurity firm McAfee unveiled new AI technology designed to protect consumers from the rising threat of deepfake audio scams and disinformation.
Dubbed "Project Mockingbird," the system uses advanced AI models to analyze videos and discern whether the audio has likely been manipulated with deepfake technology. Early tests show McAfee's detection system is over 90% accurate in identifying fake audio that has been artificially generated to impersonate real people.
The company highlighted the need to defend against the growing misuse of generative AI tools that make it relatively easy for criminals to create convincing fake audio. These convincing "vocal deepfakes" can then be used in targeted scams, like impersonating a family member asking for money, or to spread disinformation by altering genuine footage of public figures.
According to McAfee CTO Steve Grobman, Mockingbird works by running the audio through contextual, behavioral and categorical detection models. This multilayered AI approach provides unmatched capabilities in distinguishing authentic audio from vocal deepfakes.
Grobman said the technology will give consumers a valuable tool to assess the likelihood of malicious fakery, much like a weather forecast. "So, much like a 70% chance of rain helps you plan your day, our technology equips you with insights to make educated decisions about whether content is what it appears to be," he stated.
The company positioned the solution as the next evolution in using AI protect online privacy and identity. McAfee Labs developed the system in anticipation of the expanding use of cheapfakes and vocal deepfakes by cybercriminals.
Early demos of Mockingbird are being showcased by McAfee at CES 2024 this week. The unveiling comes amid heightened public concerns over advanced deepfakes. A recent McAfee survey found 84% of Americans worry about how deepfakes will be used this year, with top concerns being election interference and scams.
By leveraging AI to empower the public to identify vocal deepfakes, McAfee aims to curb the potential misuse while still allowing for innovation in generative AI technology. Project Mockingbird represents a major step toward building comprehensive safeguards against AI disinformation. However, it's crucial to recognize the inherent challenges in this field.
Detecting AI-generated content, particularly deepfakes, is notoriously difficult. The MIT Media Lab's project Detect DeepFakes highlights the complexities involved in distinguishing real from AI-manipulated media. There's no single tell-tale sign to spot a fake, but rather a multitude of subtle signs that need to be identified. This complexity is underscored by the Deepfake Detection Challenge (DFDC), spearheaded by major tech companies and academic institutions, which aimed to spur innovation in detecting deepfakes.
While McAfee's says Project Mockingbird boasts a high accuracy rate, in reality, the struggle to detect deepfakes effectively is ongoing. The evolving nature of AI-generated content means that tools like Project Mockingbird are essential, but they are part of a larger, ongoing effort to combat digital deception. As the AI landscape continues to advance, the race between creating and detecting deepfakes will undoubtedly remain a challenge.