ElevenLabs found an uptick in “voice cloning misuse cases” during its recent beta; 4chan users made deepfake voices of Joe Rogan, Ben Shapiro, and Emma Watson
how did they not see this coming? https://www.vice.com/... @emwatsonupdates : Great! Can't trust anything we see or hear now thanks to deepfakes. Be careful. https://twitter.com/... Joseph Cox / @josephfcox : Here are some examples of AI-generated voices on 4chan created over the weekend. Shows that with the rapidly decreasing barrier of entry, the trajectory of AI-generated voices is just starting. https://www.vice.com/... https://twitter.com/... See also Mediagazer
Context & Ripple Effects
The misuse ElevenLabs is reporting during its own beta was forecast years ago: Wired's 2018 examination of deepfakes flagged that cheap AI-generated audio could be weaponized against elections long before consumer cloning tools shipped. What changed with this beta is the barrier to entry — Joseph Cox documented 4chan users producing convincing clones of Joe Rogan, Ben Shapiro, and Emma Watson over a single weekend.
ElevenLabs sits at the center of the arc that followed: its tools were later used by a WSJ columnist to build a clone that defeated her bank's voice biometric system, and The Atlantic's profile found the founders themselves seemingly unprepared for how their tech reshapes public discourse. The company's growth story — including investor talks at roughly double its February valuation — now runs directly into this abuse record.
First-order effects
- ElevenLabs must police its own beta in real time: the uptick in reported misuse cases lands on the same startup courting investors at a sharply higher valuation, making moderation capability part of its commercial pitch.
- Joe Rogan, Ben Shapiro, and Emma Watson become early proof that any public figure's voice can be replicated without consent — the cloned audio spread on 4chan within days of the tool being available.
Second-order effects
- Voice biometrics lose credibility as an authentication factor: once a cloned voice can pass a bank's checks, financial institutions face pressure to move beyond voice-only verification.
- Competitors are forced into access-control arms races — OpenAI's later Voice Engine launch restricted synthesis to a small set of partners, a containment model shaped by exactly the open-abuse dynamic ElevenLabs' beta exposed.
Third-order effects
- If cloning keeps outrunning detection, the industry converges on a trust layer for synthetic media — provenance signals, watermarking, and platform-level verification — because takedowns alone demonstrably lag behind generation speed.
- Regulators inherit a ready-made case file: the election-risk scenario laid out in 2018 plus real impersonations of named public figures gives lawmakers concrete evidence for mandating disclosure of AI-generated audio.
The trend: Synthetic voice is following the deepfake trajectory predicted years ago — capability commoditizing faster than safeguards, pushing the burden onto detection infrastructure and eventual regulation rather than vendor restraint.