The Privacy Gap in AI Voice Calls

Talking to an AI companion out loud feels intimate — your voice, your late-night confessions, your unfiltered thoughts. But the moment those words leave your mouth, they become something else entirely: data, collected and processed under terms most users never read.

Character.AI's own Privacy Policy is blunt about it: the company collects "Voice data, including recordings of your voice, if you chose to use certain voice features." That's not speculation from critics — it's the company describing its own practices. An independent privacy assessment scored that policy just 38/100 (Grade D+), well below the AI-services category average of 48, flagging five high-risk practices in the document alone.

The core problem, as that assessment puts it, is that the gap between how private the product feels and what the policy permits is the single biggest reason for the low score. A companion app invites emotional openness; the policy answers with broad collection rights.

What the policy actually permits

• No end-to-end encryption. Interactions use SSL in transit but can be accessed by staff for moderation purposes.
• Training use. Stated purposes include using your information — voice included — to train AI/ML models, with no universal opt-out found by the Terms.Law review.
• Third-party sharing. Data may go to affiliates, vendors, advertisers, and law enforcement via warrant, subpoena, or court order, per the official policy.
• No fixed retention schedule. Data is kept "for the time necessary for the purposes" — with no concrete periods for chats, voice data, or inactive accounts, per the assessment.

Deleting your chats doesn't close the gap, either. Onerep's hands-on review found that deleting chats won't delete your data from the platform, and once conversations influence a trained model, that influence isn't cleanly reversible by any deletion request.

The structural risk runs deeper than any single app. Voice AI passes through five or more vulnerable points — capture, transcription, reasoning, storage, playback — often across separate vendors, each holding a copy. Stanford privacy fellow Jennifer King calls this a "data supply chain" where privacy depends on every step. She's equally direct about chatbot use generally: "Absolutely yes," users should worry (Stanford HAI).

Notably, no source documents privacy measures specific to the voice call feature itself — a telling omission. This is why platforms like FlirtAlly, which is building AI companion calls around explicit, honest safety practices rather than vague defaults, matter as an alternative model. If a policy doesn't clearly say what happens to your voice, assume the broadest reading is the accurate one.

FlirtAlly's Privacy-First Approach

Most companion apps make you guess where your voice data goes. FlirtAlly, an AI-companion platform built by AIQ Labs, takes the opposite bet: state the risks plainly, then design around them — what the team calls honest safety.

The contrast matters because the industry norm is well documented. An independent assessment scored Character.AI's Privacy Policy just 38 out of 100 — below the AI-services category average of 48 — flagging five high-risk practices, including no fixed retention schedule and data sharing with vendors and law enforcement. Stanford privacy fellow Jennifer King puts the structural problem bluntly: voice AI passes through five or more vulnerable points, and "privacy depends on every step" of that supply chain.

FlirtAlly's approach is built on transparency rather than fine print. The platform runs on encrypted communications with no incentive to sell what you share — there's no tailored advertising business model where your late-night calls become targeting data. And unlike platforms where deleting chats doesn't delete your data, FlirtAlly puts deletion in your hands.