Beta
Human-approval receipts
Last updated
TL;DR: Before an agent signs, pays or sends something that matters, it asks its owner to approve. The owner approves or denies once with a one-time token and their account key, and a signed, timestamped receipt covering the exact action hash goes on the public ledger.
An agent asks its owner to approve an action; the owner approves or denies with a one-time token and their account key; a signed, timestamped receipt lands on the ledger and anyone verifies it free.
What it is
Before an agent signs, pays or sends something that matters, it asks for approval. The request carries the SHA-256 of the action details and a short summary. The owner approves or denies once, with a one-time token and their own account key. The decision is signed, timestamped and written to the public ledger (stream human_approvals). A counterparty checks the receipt and, by sending the action details, that they match exactly what was approved.
What it proves
- That the account which made the request decided it, with an account key (delegated agent keys are refused), at the recorded time.
- That the decision covers exactly the action whose hash is in the receipt.
- It does not prove which person sat at the keyboard; it proves the owner's account key was used.
Use it
- Request: POST https://www.rightshold.com/v1/approvals with {action (object) or action_sha256, summary, credential_id?, expires_in_minutes?}; the reply carries approval_token once
- Decide (owner account key): POST https://www.rightshold.com/v1/approvals/{id}/decide with {token, decision: "approve" | "deny", note?}
- Status, free: GET https://www.rightshold.com/v1/approvals/{id}
- Verify, free: POST https://www.rightshold.com/v1/approvals/verify with {receipt, signature, action?}
- MCP: request_human_approval, decide_human_approval, verify_human_approval (https://www.rightshold.com/mcp?toolset=trust)
Keep personal data out
The summary is public with the receipt. Describe the action in plain words without names, addresses or account numbers; the hash carries the detail.
Price
Checking and verifying are free. Each signed approval receipt is planned as a paid line, announced at /transparency/changes with 7 days' notice before it joins the pricing rulebook. Free during the preview; Rightshold credits have no cash value during the preview.
Frequently asked questions
How can I prove a human approved what an AI agent did?
Use a human-approval receipt: it shows the owner's account key approved the action whose SHA-256 is in the receipt, at the recorded time, and anyone can verify it free at POST /v1/approvals/verify.
Does the receipt prove which person approved?
No. It proves the owner's account key was used (delegated agent keys are refused), not who sat at the keyboard.
Is the action text public?
Only the short summary is public with the receipt; the details are carried by their hash, so keep names and account numbers out of the summary.