- ElevenLabs offers cutting-edge speech synthesis and voice cloning technology.
- You can clone voices from just 30 seconds of voice recordings.
- There are many documented cases of controversy and misuse.
ElevenLabs was initially founded by former Google and Palantir employees, securing a huge $19 million in Series A funding. The company enables users to generate exceptionally lifelike voices from scratch or clone voices to create AI voices that can be manipulated at will.
While AI-generated voices have been around for a few years, ElevenLabs takes it to a new level. Using just 30 seconds of voice recordings, you can clone a voice into something you can manipulate to say pretty much anything.
You can probably already imagine the kinds of problems ultra-realistic voices might cause.
ElevenLabs roused controversy when users on forums like 4chan began using it to clone celebrity voices for unethical purposes, including spreading misinformation and hate speech. Deeper than that, voice cloning has been used to support scams, implicate public figures in illegal activities, and risk harmful or fraudulent fake content going viral on social media.
Misuse has caused uproar in the AI industry and raised questions about ElevenLabs’ ability to control its services.
While users are only meant to clone voices they have the rights to, it’s not something ElevenLabs can police easily.
Keir Starmer voice cloning incident in the UK
An audio clip allegedly featuring Sir Keir Starmer, the leader of the UK’s Labour Party, swearing at a staff member went viral on social media.
However, there is no evidence to support the authenticity of this clip. It has over 1.5 million views and has been widely shared prior to being widely outed as fake.
Labour sources and Conservative MPs, including UK security minister Tom Tugendhat, have stated that the clip is fake.
Slovakian election incident
A similar incident occurred in Slovakia, where an audio recording allegedly featuring the head of a political party and a prominent journalist discussing how to rig the vote was posted to Facebook.
Both individuals denied the conversation, and fact-checkers noted signs of AI manipulation in the audio.
The fake recording appeared in the scheduled pre-election media blackout, and the election itself won by a fine margin.
Omar Al Bashir incident in Sudan
In Sudan, amidst the civil war, an AI campaign on TikTok impersonated Omar al-Bashir, the country’s former leader.
This campaign, which posted’ leaked’ ex-president recordings, garnered widespread attention on TikTok and added to the confusion surrounding the war-torn country.
Investigations into the recordings suggested that voice replication software like ElevenLabs was used to mimic Bashir’s speech.
ElevenLabs’ response to the misuse
Responding to concerns and controversy, ElevenLabs has taken steps to secure its platform and guide responsible use.
They published a Voice Cloning Guide, emphasizing their technology’s legal and ethical boundaries.
While they allow uses such as caricature, parody, satire, and artistic and political speech, they strictly prohibit uses that involve fraud, hate speech, or online abuse.
The company has also stated its readiness to suspend accounts and report illegal activities in line with applicable laws.
But again, while policies and posts are one thing, locating illicit use cases is non-straightforward.
Balancing innovation and ethics
The case of ElevenLabs highlights the dual-edged nature of AI.
We’ve seen many examples of the technology’s potential for innovation being tempered by ethical and security challenges.
Deep fakes – AI-generated fake images replicating real content – are certainly one of the technology’s most pressing risks.
While ElevenLabs illustrates this in the context of voice cloning, the music industry is the center of the debate surrounding AI’s ability to copy artists’ voices.
We’ve already witnessed the success of AI-generated songs like ‘Heart on My Sleeve,’ produced using lyrics from Drake and The Weeknd without either’s permission. YouTube recently adjusted its policies to combat such types of content, following suit with other platforms like Meta and TikTok.
In any case, with more AI developers investing in audio manipulation tools, as indicated by Google’s Lyria models, the plot is set to thicken.