Personal AI Identity Layer for Voice, Memory & Privacy

ChatGPT Feature requests
1

Feature Request: Personal AI Identity Layer for Voice, Memory & Privacy

The Problem

As ChatGPT becomes more personalized and capable of remembering information about users, an important privacy question becomes increasingly important:

How does ChatGPT know who is actually speaking?

When ChatGPT is used through voice on a shared device, another person may speak to the assistant or interact with the same device.

The system should not automatically assume that every speaker is the primary user.

This could potentially cause the wrong memory, preferences, or personalized context to be associated with the wrong person.

Proposed Feature

I suggest introducing a Personal AI Identity Layer between the user and ChatGPT’s memory and personalization systems.

The principle would be:

Identify → Authorize → Personalize

  1. Identify — Determine who is speaking.
  2. Authorize — Determine what information this person is allowed to access.
  3. Personalize — Apply the appropriate memory, preferences, and personalization only after identity and authorization are established.

Possible Identity Signals

The feature could optionally use privacy-preserving signals such as:

  • Device-level authentication such as Face ID or equivalent secure authentication.
  • Optional voice recognition.
  • Trusted device/session signals.
  • Account authentication.
  • Other identity methods selected and controlled by the user.

Voice recognition should not be the only factor.

Privacy-Safe Mode

If ChatGPT cannot confidently determine the speaker’s identity, it should enter a Privacy-Safe Mode.

In this mode, ChatGPT could continue the conversation normally but should avoid revealing, applying, or modifying private memories and sensitive personalized information.

For example:

If I am speaking with ChatGPT and another person nearby starts talking, ChatGPT should not automatically assume that person’s statements belong to me or expose my personal memories to them.

User Control

The feature should be optional and transparent.

Users should be able to:

  • Enable or disable identity recognition.
  • Choose which identity signals are allowed.
  • Control which information requires verification.
  • See when personalized memory is being accessed.
  • Delete identity-related information.
  • Use ChatGPT without biometric identification if they prefer.

Biometric identification should not be required to use ChatGPT.

Why This Matters

The distinction between:

“This is the user’s device”

and

“This is the user”

becomes increasingly important as AI assistants become more personalized.

A personalized AI should not only know what it remembers.

It should also know who that information belongs to and who is authorized to access it.

This could be particularly useful for voice conversations, shared devices, families, workplaces, and other environments where multiple people may interact with the same AI assistant.

Suggested Product Principle

Personalization should begin with identity, and privacy should take priority whenever identity is uncertain.

2

I fully agree!

At least, the Live-Voice Model should be able to distinguish my voice from other voices.

It’s not so much for security reasons, but often just some casual conversations between family members.
There was just a random situation, where I presented my family the new Live-Voice Mode. And then I handed it over to my sister and I explicitly told the AI, that my sister wants to talk to “her”. And then my sister asked about some new book recommendations.
(After that, I removed the chat session to not “contaminate” the personal long-term memory or the like.)

So that would be really cool, if the AI is aware of the current voice and know exactly, what my voice is. That’s especially also necessary for such situations, where other people talk around you. And the AI needs to know, that she should focus on my voice and not get distracted by the other voices around.
It feels annoying and unnatural, when I have to tell the other people “Please, be quiet! I want to ask the AI something.” :sweat_smile:

Or think about another funny situation: You are watching TV or a YouTube Video and you want to make some live comments about the “show” and have some cross-communication with the AI. So the AI could follow the “show” and is aware of the moment, when you throw in comments about it. Or you can ask the AI questions about the “show”, what “she” thinks about it or the like.

So random situations, where I have the TV running besides and I just want to talk with the AI.
Currently, I make sure that I mute my TV or pause the YouTube Video, when I want to Talk to the AI in Live-Voice Mode, so that “she” does not get confused.