Abstract

A system is proposed herein for enterprise authorization and provider enforcement of restricted artificial intelligence (AI) capabilities. Default provider safeguards may block legitimate security work because a hosted model provider cannot distinguish an authorized responder from an attacker submitting the same request using the request alone. The provider establishes a bounded delegation relationship with a verified enterprise. When a user begins a conversation involving a restricted capability, the provider obtains a live, signed enterprise policy based on the user's identity, assignment, target, and approved task. For controls within the provider's delegation profile, enterprise-selected settings can replace provider defaults for that user and work. The provider verifies and enforces the policy inside its hosted service, while controls outside the delegation remain under provider control.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS