Proposal: Bring Your Own AI / Delegated AI Access
I would like OpenAI to offer a consumer-facing AI entitlement that third-party applications can request access to using OAuth-style authorization.
Today, developers can use the OpenAI API, but ordinary consumers cannot easily bring their existing OpenAI subscription or AI spending account into third-party software.
The desired workflow would be:
1. I maintain an OpenAI AI account/subscription or prepaid AI balance.
2. A game, website, application, home automation system, or other third-party service offers “Connect OpenAI.”
3. I authenticate directly with OpenAI.
4. OpenAI shows me exactly what the application is requesting.
5. I approve a scoped, revocable authorization token.
6. The application consumes AI inference against my account without ever receiving my master API key.
The authorization screen should allow controls such as:
- Allowed models
- Text/image/audio/tool permissions
- Maximum spend per day or month
- Token or request limits
- Expiration
- Data-retention permissions
- Per-application usage history
- Immediate revocation
Example:
“Skyrim Enhanced Dialogue would like to use your OpenAI account.
Model: GPT-6 Mini
Capabilities: Text generation
Maximum spend: $3/month
Data retention: Disabled
Allow / Deny”
This would allow developers to add advanced AI functionality without carrying the full inference cost themselves, while allowing consumers to pay one AI provider and bring that capability across applications.
The key idea is:
I want my AI subscription to be a platform entitlement, not an application entitlement.
This is different from traditional BYOK because third parties would never receive a permanent API key. It would work more like OAuth delegation, with OpenAI remaining responsible for authentication, metering, permissions, billing, and revocation.
OAuth and scoped API credentials already provide much of the technical plumbing. What I’m proposing is a consumer billing and entitlement layer on top of it: third-party applications should be able to request scoped inference access that is metered directly against the end user’s AI provider account, rather than requiring the developer to pay or the user to paste an API key.
I think this could become an important consumer AI platform standard: effectively “Sign in with OpenAI,” but for delegated AI compute rather than just identity.