A Consumer Reports investigation revealed that most leading artificial intelligence voice cloning programs lack adequate safeguards, allowing individuals to nonconsensually impersonate others. In recent years, voice cloning AI technology has advanced significantly, enabling many services to convincingly replicate a person's speech patterns with just a few seconds of audio. A notable incident occurred during last year’s Democratic primaries when robocalls featuring a fake Joe Biden flooded voters' phones, advising them not to vote. The political consultant behind this scheme received a $6 million fine, and the Federal Communications Commission has since prohibited AI-generated robocalls. A recent survey of six major publicly available AI voice cloning tools identified that five of them have easily circumvented security measures, making it simple to clone voices without consent. Deepfake audio detection software often finds it challenging to distinguish between real and synthetic voices. Generative AI, which imitates human attributes such as voice, writing, and appearance, is a rapidly evolving field with minimal federal regulation. Most safety and ethical guidelines within the industry are self-imposed. President Biden included some safety measures in his 2023 executive order on AI, but those were revoked by President Trump upon taking office. Voice cloning technology operates by capturing an audio sample of a person's voice and producing a synthetic audio file based on that sample. Without proper safeguards, anyone can create an account, upload audio from a platform like TikTok or YouTube, and have the service replicate that person’s voice. Four tools—ElevenLabs, Speechify, PlayHT, and Lovo—simply require users to check a box confirming that the person whose voice is being cloned has authorized it.

Resemble AI has a slightly stricter requirement, necessitating real-time audio recording. However, Consumer Reports managed to bypass this by playing a previously recorded audio file from a computer. The only service with a fairly effective safeguard is Descript, which mandates that the person wanting to clone a voice records a specific consent statement, making it harder to falsify compared to using other services. All six tools can be accessed publicly on their websites, with ElevenLabs and Resemble AI charging $5 and $1, respectively, to create a custom voice clone, while the others are free to use. Some of these companies acknowledge the potential for misuse of their products and claim to have implemented strong safeguards against creating deepfakes or facilitating voice impersonation. A Resemble AI spokesperson stated to NBC News, “We recognize the potential for misuse of this powerful tool and have implemented robust safeguards to prevent the creation of deepfakes and protect against voice impersonation. ” There are legitimate applications for AI voice cloning, such as assisting individuals with disabilities and generating audio translations in multiple languages. However, the potential for misuse is significant, according to Sarah Myers West, co-executive director of the AI Now Institute, a think tank examining AI policy implications. She told NBC News, “This could obviously be used for fraud, scams, and disinformation, notably impersonating institutional figures. ” Research on the prevalence of AI in audio-based scams is limited. In grandparent scams, for instance, a criminal contacts someone claiming there is an emergency involving a family member, such as a kidnapping or injury. The Federal Trade Commission has alerted the public about the possible use of AI in these schemes, although such scams existed before the advent of this technology.