In a recent incident that gained significant attention, over 130 million individuals viewed a video featuring a deepfake recording of Kamala Harris, reshared by entrepreneur Elon Musk on his social media platform. Although the original creator had classified the video as a parody on YouTube, Musk neglected to include this descriptor, potentially leading some viewers to mistake the audio as genuine rather than a manipulated deepfake or voice clone of Kamala Harris.
The dissemination of this deepfake recording underscores the potential threat posed by audio deepfakes as the U.S. presidential and local municipal elections approach in November. Rahul Sood, the chief product officer at Pindrop Security, an IT security firm, noted that the existing AI safety standards may not be adequate to safeguard the integrity of elections in light of such incidents.
The growth of voice cloning and deepfake technology has experienced exponential growth in recent years, particularly with the emergence of generative AI and LLMs. Such technology has already impacted several elections globally, such as in Slovakia, where a fabricated audio clip of a prominent candidate alleging election rigging circulated.
According to Sood, the proliferation of generative AI has paved the way for a synthetic voice market that enables users to create voice-cloning audio for as little as $1 per month. While detecting fabricated audio was more manageable in the past, the swift evolution of voice cloning tools has made it increasingly accessible for individuals to create deepfakes online.
The concept of the “uncanny valley” has become pertinent in the realm of synthetically generated audio, with Sood highlighting the challenge of discerning between authentic and artificial audio content. The ease with which deepfake technology can be employed underscores the potential risks posed to election integrity.
Instances of deepfakes during elections have already emerged, with a notable incident involving a deepfake audio of President Joe Biden used in a robocall to dissuade New Hampshire voters from participating in an election. The use of voice cloning technology to generate such deepfakes underscores the need for enhanced safeguards to prevent the spread of fabricated content.
Moreover, the partial deepfake involving Vice President Harris reveals the shortcomings of existing protective measures. Issues such as consent requirements for voice cloning platforms, lack of standardized watermarking practices, and inadequate third-party testing of voice consent technology underscore the need for more robust regulations and procedures.
Looking ahead to the 2024 election, experts anticipate the pervasive impact of deepfake technology on the electoral process. Lisa Martin from the Futurum Group highlights concerns regarding the credibility of audio deepfakes and their potential influence on voter perceptions.
The responsibility of tech vendors in addressing the threat of deepfakes is paramount, with the need for increased awareness, detection tools, and regulatory measures. Efforts from vendors like Deep Media AI and Pindrop to develop deepfake detection technologies, alongside government initiatives such as proposed bills to criminalize AI-generated content aimed at influencing elections, signal a collective effort to combat deceptive practices.
As the evolution of deepfake technology continues to pose challenges to election integrity, the collaboration between tech vendors, government authorities, and the public is crucial in safeguarding the democratic process from malicious manipulation. The ongoing discourse surrounding deepfakes underscores the need for proactive measures to mitigate their impact on electoral outcomes.