Abstract
This disclosure describes a method by which an AI agent can act on a person's or organization's accounts without ever holding the passwords, keys, tokens or session cookies that give access to them. Credentials are kept in a credential proxy on an independent safety plane, with its own processor, firmware, storage and power, which the agent, its model and the host operating system cannot read. When the agent needs to act on an account, it submits a structured action record to the governance point. If the action is admitted, the governance point issues a single-use authorization bound to the hash of that exact request. The proxy checks the authorization, attaches or uses the credential for that request only, sends it, and returns the response to the agent with all secrets, session tokens and credential-bearing headers removed. For high-risk actions the proxy requires the owner's live presence before it uses the credential. When the governor level rises, the proxy narrows or stops all credential use at once. Every use is recorded in a tamper-evident ledger. A confused, manipulated or compromised agent therefore has no credential to leak, copy or misuse, and can only cause the specific actions that governance admits.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Regulator™: Credential Proxy on an Independent Safety Plane, so that AI Agents Act on Logins and Keys They Never Hold", Technical Disclosure Commons, (October 13, 2026)
https://www.tdcommons.org/dpubs_series/12172