On March 19, 2026, Biometric Update highlighted a hard truth about many AI companion apps: your most emotionally intimate chat data may not be stored as securely as many of us assume. The article summarized an Oversecured report claiming that 17 popular Android AI companion apps, representing more than 150 million installs, contained 14 critical flaws and 311 high-severity issues.
People use AI companion apps to talk about loneliness, sexuality, conflict, grief, and dependency. These apps collect exactly the kind of data that can lead to life-altering privacy failures if it is not properly protected.
Aicrest takes a distinctly different approach to privacy. Its privacy-first architecture puts it far ahead of cloud-based AI companion apps.
With Aicrest, your speech data and transcribed messages never leave your device, keeping them under your control. The app uses local language models to generate responses directly on iPhone and other Apple devices. This architecture dramatically reduces the risk of intimate conversation data falling into the wrong hands somewhere on the internet. On-device, your chat data is stored inside the app’s sandboxed environment and protected by operating system security mechanisms such as Face ID. The article’s warning is fundamentally about the concentration of sensitive data in remote systems and weak app-layer security. On-device AI companions minimize that risk, even if they cannot eliminate it entirely.
Local language models are not a silver bullet. Some features are still difficult to deliver fully on-device, most notably vocalizing AI companion responses in the voice of your choice. Aicrest sends text-to-speech requests to cloud voice providers such as ElevenLabs and Hume to make conversations with AI companions feel more natural and immersive.
This is where the Aicrest backend proxy becomes important. It sends text-to-speech requests to voice providers on behalf of Aicrest users and returns the resulting audio to the correct user. As a result, voice providers have no direct way of knowing which request originated from which person.
Beyond its privacy-first architecture, Aicrest also covers the basics of strong privacy hygiene. The app makes data retention and deletion available as product features. Users can set chat retention windows, export their companion data, import it back, or delete local data entirely. That directly addresses one of the article’s underlying concerns: companies often retain intimate data indefinitely, making it more vulnerable to breaches.
Aicrest is also built with explicit privacy controls instead of burying them in policy text. It separates consent into concrete categories such as AI interaction, cloud processing, voice processing, location context, link lookup, and cross-device sync. That is a much better model than treating “agree to everything” as a single switch. It creates a clearer privacy boundary and keeps sensitive features off unless the user has explicitly granted permission.
Aicrest is a native iOS app that lets you interact with AI companions using on-device response generation, anonymized response vocalization, and clear consent segmentation. Together, these choices create a much healthier security posture than a thin chat frontend glued to cloud APIs.
The article is a warning to the entire app category. As AI companion app developers, we all need to work on delivering stronger privacy protections so people can speak more openly with AI companions and have more personal, immersive conversations. That matters because AI companions can offer something uniquely valuable: a space to be heard without judgment and, in some cases, a way to help people rebuild confidence and engage more fully with the real world.