AI Agent Authority Mandate Grant (US) by OCS
This document establishes a legal framework for a business to authorize an AI agent operated by a third-party vendor to act on its behalf. It defines the AI agent's authority through specific spend caps, permitted transaction categories, and mandatory human approval thresholds.
AI AGENT AUTHORITY AND MANDATE GRANT
This AI Agent Authority and Mandate Grant (this “Mandate”) is made effective as of [effective date] (the “Effective Date”), by and between:
(1) [principal legal name], a [principal entity type] organized under the laws of [principal state of organization] with its principal place of business at [principal address] (the “Principal”); and
(2) [agent operator legal name], a [agent operator entity type] organized under the laws of [agent operator state of organization] with its principal place of business at [agent operator address] (the “Agent Operator”).
The Principal and the Agent Operator may be referred to individually as a “Party” and together as the “Parties”.
Note: Who signs this. This Mandate assumes the Principal and the Agent Operator are different entities: a business authorizing an AI agent that a vendor builds and operates for it. If you built or configured the agent yourself, you are both parties, and this form should not be signed as it stands. An indemnity you give yourself does nothing, the Agent Operator warranties in Section 10.2 become warranties you give to yourself, and the independent contractor language in Section 15.7 does not describe the arrangement. Sections 3 through 9 and Exhibit A still do useful work in that situation if you adopt them as an internal authorization, approved by your board or managing member and signed by the officer responsible for the deployment. What you lose is the risk allocation in Section 11, which was never available to you in the first place.
WHEREAS, the Agent Operator operates the AI Agent, an automated system that incorporates AI Technology and is capable of initiating, negotiating, and executing transactions and other legally significant acts without human review of each individual action;
WHEREAS, the Principal wishes to authorize the AI Agent to act on the Principal's behalf, subject to the limits of authority set out in this Mandate and in the Authority Schedule at Exhibit A, and the Agent Operator agrees to operate the AI Agent within those limits;
WHEREAS, the Parties intend that an executed copy of this Mandate, including Exhibit A, may be presented to Third Parties as evidence of the AI Agent's authority to act for the Principal and of the limits of that authority;
NOW, THEREFORE, in consideration of the mutual covenants set out in this Mandate, and for other good and valuable consideration, the receipt and sufficiency of which are acknowledged, the Parties agree as follows:
1. DEFINITIONS
This is a preview. The full template is free on GitLaw.
5.0 out of 5 on Google
Read reviewsAs seen in
About this template
What is this template?
This document is a formal delegation of authority specifically for autonomous software systems known as AI agents. It is not a software service agreement; it is a mandate that establishes the legal attribution of the agent's acts to the principal under US electronic commerce laws. It contains technical enforcement obligations for the operator and a detailed 'Decision Log' requirement for every automated act.
When should you use it?
Use this when a business wants to allow an AI system to enter into binding contracts with third parties without a human reviewing every individual action. If the business is building and operating the agent internally rather than using a vendor, an internal corporate resolution should be used instead.
What's inside
| Clause | Name | What it does |
|---|---|---|
| 2.2 | Electronic Agent Designation; Attribution | Designates the AI Agent as an 'electronic agent' under UETA Section 14 and E-SIGN Section 7001(h) to bind the Principal to transactions. |
| 3.2 | Cumulative Limits | Conditioning authority on satisfying per-transaction spend caps, aggregate spend caps, and permitted counterparty types listed in Exhibit A. |
| 4.1 | Actions Never Permitted | Excludes 12 specific 'Prohibited Actions' from the AI Agent's authority, including real property leases, guarantees, and hiring employees. |
| 5 | Human Approval Thresholds | Requires express prior approval from an Authorized Approver through a designated channel for actions exceeding specific triggers in Exhibit A. |
| 8.2 | Disablement | Obliges the Agent Operator to technically disable AI Agent capabilities within a specific number of hours after receipt of a revocation notice. |
| 9.1 | Decision Log | Mandates the Agent Operator maintain a human-readable log of every act, including inputs relied on and transaction values evaluated. |
| 11.2 | Acts Exceeding the Mandate Are the Agent Operator's | Requires the Agent Operator to indemnify the Principal for losses arising from acts exceeding the Scope of Authority or Prohibited Actions. |
| 11.4 | Cap on the Agent Operator's Mandate-Breach Liability | Provides a choice between uncapped liability (Option A) or a specific liability cap amount or formula (Option B). |
Who it's for
- a US company authorizing an AI agent operated by a third-party vendor to execute commercial transactions
- an AI system operator requiring express delegation of authority from a principal to manage automated contracting
- a principal seeking to limit automated agency through per-transaction and aggregate spend caps
- parties intending to deploy automated systems under the UNCITRAL Model Law on Automated Contracting
How long it runs and how it's signed
Law it's drafted under
Frequently asked questions
A template isn't binding on its own - like any contract, it becomes binding once it's properly completed and signed. Templates in our curated library are professionally drafted for US or UK law; review any template before you sign it.
Yes. Chat with GitLaw to edit any section, or make changes directly in the editor.
Yes, read about team plans here.
Describe what you need in the chat and GitLaw will draft it for you.
Templates in our curated library are professionally drafted for US or UK law. The wider library comes from the GitLaw community and public sources - a solid starting point, but check any template fits your situation before you rely on it.
Mostly US and UK law. Some templates use general commercial terms that work across jurisdictions, and many note which law they're written for.
It depends on the situation. Templates work well for routine business agreements. For anything involving significant money, complex IP, employment, or areas you're unsure about, it's worth getting professional advice before you sign. GitLaw provides templates and tools, not legal advice.
Open any template in GitLaw and describe the change you want in the chat — 'make clause 4 mutual' or 'add a 30-day notice period', for example. GitLaw drafts the revised language and shows it as a suggested edit. You accept, reject, or keep editing from there.
Yes. Upload a Word, PDF, or Markdown file and GitLaw will open it in the editor. You can review, edit, or chat with GitLaw about it the same way you would with any template from the library.
Trusted by 5,000+ businesses
From template to signed, in one place
Every template opens in an editor with an AI agent alongside it.
Open
Pick a template and open it. Nothing to download, and no credit card to start.
Free to open
Edit with AI
Describe your situation in chat and the agent adapts the wording, clause by clause.
Tracked changes you can review
Send and sign
Share it for negotiation, then collect signatures without leaving GitLaw.
eSign included
Built for your legal work,
with practicing lawyers
Trained on 5,500+ clauses and specialist areas of law. Built with a standards committee of independent lawyers.
As seen in
Ready to get started?
No sales calls, no credit card. Just chat with GitLaw.
GitLaw provides templates and tools, not legal advice. Templates are a starting point, not a substitute for advice on your situation - for anything significant, speak to a qualified lawyer.