Kalypta Mutes Your AI Overlords by Turning Your Voice Into Gibberish
Finally, a way to talk without your computer eavesdropping on your every word. Deveillance just dropped Kalypta, a clever piece of software that makes your voice perfectly clear to humans but absolute nonsense to the AI models hungry for your data.
The app works by running a local model that injects specific audio perturbations into your outgoing stream in real-time. By subtly messing with the sound waves, Kalypta makes speech recognition systems like OpenAI's Whisper and Nvidia's Canary hallucinate wildly. Internal tests show these models failing to transcribe roughly two out of every three words while your human colleague on the other end hears you perfectly fine.
Protecting private calls from automatic transcription might seem like a paranoid dream, but the startup is betting on it. While tech giants could theoretically retrain their models to ignore these distortions, it would require a massive effort to update their training datasets just to hear your morning stand-up. The software is currently in beta for macOS, and Deveillance is already known for their Spectre I hardware that uses acoustic signals to jam nearby microphones.
This is effectively a digital arms race where users are desperately trying to build noise-canceling headphones for their privacy. Whether this becomes a standard tool for every corporate meeting or just another obscure privacy toy for the terminally paranoid remains to be seen. The real question is how long until the AI companies patch this and render our voice-cloaking efforts as useless as a screen door on a submarine.
Source: Deveillance
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.