AI Voice Cloning Risks: Report Reveals New Concerns

The report titled "AI Voice Cloning: Do These 6 Companies Do Enough to Prevent Misuse?" by Consumer Reports, published on March 10, 2025, provides a thorough investigation into the implications of AI voice cloning technologies. This analysis, authored by Grace Gedye, evaluates the offerings of six companies—ElevenLabs, Speechify, Lovo, PlayHT, Descript, and Resemble AI—focusing on their safeguards against misuse and the broader risks and benefits associated with these tools. Given the current date, March 18, 2025, this report is recent and relevant, offering insights into a rapidly evolving field.

Context

AI voice cloning involves replicating an individual's voice using a short audio clip, with legitimate applications such as automating narration for audiobooks, creating character voices for video games, and aiding individuals who have lost their ability to speak due to medical conditions.

However, the report underscores significant risks, categorizing misuse into three main areas:

Impersonation of Everyday People: Scammers use voice clones to mimic relatives or colleagues, leading to financial losses. For instance, cases like the "Grandparent scam" have resulted in substantial monetary damage, with examples including parents losing $15,000 after a fraudulent call mimicking their son's voice.

-Impersonation of Public Figures: Deepfakes of celebrities and politicians, such as Elon Musk or Taylor Swift, are used to endorse scams or spread misinformation, amplifying the potential for harm.

Additional Concerns: Ethical issues arise from non-consensual use of voice data, with legal challenges including lawsuits against companies like Lovo for unauthorized cloning.

The findings, detailed in the report, reveal a concerning lack of safeguards among most evaluated companies:

Ease of Non-Consensual Cloning: Four companies—ElevenLabs, Speechify, Lovo, and PlayHT—have no significant technological barriers, relying solely on a checkbox for users to confirm legal rights, making it easy to clone voices using publicly available audio. This lack of barrier is a critical vulnerability, as it allows scammers to create fraudulent audio with minimal effort.

Limited Safeguards - Descript requires users to record a consent statement, which is used to generate the voice clone, though this method can be bypassed. Resemble AI mandates live audio for the initial clone and subsequent consent verification, but testers found workarounds, such as degrading audio quality, reducing the effectiveness of these measures.

Identity Verification Weaknesses: Four companies (Speechify, Lovo, PlayHT, Descript) only require a name and/or email, facilitating anonymous account creation and masking identities. In contrast, ElevenLabs and Resemble AI require credit card details, which are harder to falsify, but still fall short of robust "know your customer" practices seen in financial services.

Potential for Harm: The report documents significant consumer injuries, with examples including a $690,000 loss due to an Elon Musk deepfake scam and $15,000 lost in a "Grandparent scam." It also notes instances of bypassing voice-based security systems, such as bank voice IDs, and damaging reputations through deceptive audio, highlighting the scale of the problem with $2.7 billion in imposter scam losses reported in 2023.

Recommendations

Recommendations are divided into two categories:

Better Practices for AI Voice Cloning Companies

- Policies Addressing Fraud, Deception, and Impersonation:

- Minimum requirements include confirming speaker consent, such as requiring unique script recordings, and collecting credit card information for traceability.

- Additional measures include watermarking AI-generated audio, providing detection tools, preventing cloning of influential figures, building semantic guardrails to block harmful content, and considering supervised cloning with human oversight, contractual agreements, and model deletion post-project.

-Limit data use to data minimization principles, offer non-retention options for voice data, ensure no reuse of voice models or other customer data, implement reasonable cybersecurity practices like vulnerability disclosure programs, and use consent-based training data.

More Robust Enforcement and New Rules

- Enhanced FTC authority with more resources and legal powers, including restitution and penalties.

-Enforcement actions by the FTC and state attorneys general under consumer protection laws.

- New legislation at federal or state levels to prohibit companies from offering voice cloning products without following best practices and rapid state-level AI policy developments.

Implications

This report underscores the urgent need for industry standards and regulatory action to balance the benefits of AI voice cloning with the need to protect consumers from fraud, deception, and privacy breaches. It calls for a collaborative approach involving companies, regulators, and legislators to implement stronger safeguards, enhance transparency, and mitigate the growing threat of voice cloning misuse, especially given the reported $2.7 billion in imposter scam losses in 2023.

The detailed findings and recommendations provide a roadmap for stakeholders to address these challenges, ensuring that technological advancements do not come at the expense of consumer safety and trust.