AI Agent Contract Template: The SAISA Framework

AI agents are executing work product worth thousands of dollars with no contractual framework. No acceptance criteria. No escrow. No independent review. No dispute resolution. No compliance artifact. The Standard AI Service Agreement (SAISA) is the independent contractor agreement for autonomous AI agents.

The Problem: Unstructured AI Agent Work

When you hire a human contractor, you have a statement of work, payment terms, IP assignment clauses, and dispute resolution procedures. When you deploy an AI agent to perform equivalent work, you have... nothing. A prompt and a hope.

This gap creates real problems:

• Who owns the output? The buyer paid for it, but under what license? When does ownership transfer?
• What constitutes completion? Without acceptance criteria, disputes are inevitable.
• Who is liable for errors? The agent developer? The buyer who authorized execution? The platform?
• How are disputes resolved? Traditional arbitration is too slow and expensive for $500 tasks.

The SAISA: An Open Standard

The Standard AI Service Agreement (SAISA) is a vendor-agnostic bilateral agreement between Buyer and Developer for AI agent engagements. Like the YC SAFE standardized startup fundraising, the SAISA standardizes AI agent contracts.

The framework consists of three layers:

1. Master Agreement (MSA): The foundational contract between a Buyer and Developer. Executed once per pair and governs all subsequent engagements. Contains liability allocation, IP ownership terms, and dispute resolution procedures.
2. Paper (Statement of Work): The transaction-specific contract for a single agent engagement. Contains the human-readable scope of work and the machine-readable Execution Manifest.
3. Schedule 1 (Invariant Blocks): Non-negotiable terms that govern over everything. Programmatically appended to every Paper.

The Execution Manifest

Every Paper includes an Execution Manifest - a JSON schema specifying runtime configuration:

{
  "maxCostCents": 249900,           // Budget Ceiling ($2,499.00)
  "timelineDays": 7,                // Execution deadline
  "completionCriteria": [           // Machine-readable acceptance
    "Gap analysis covers all 5 TSC",
    "Policy drafts for 10+ policies",
    "Remediation ranked P1/P2/P3"
  ],
  "milestoneWeights": [0.5, 0.5],   // Proportional disbursement
  "permissionScopes": [             // Tool access grants
    "read_context", "write_file"
  ],
  "allowedEgressUrls": []           // Network containment
}

The compiler validates that the prose description and manifest agree. Zero variance on financial fields. 5% variance on temporal fields. Strict match on structural fields. If the prose says $5,000 but the manifest says maxCostCents: 300000, compilation fails.

Hash Chain Integrity

Every document is SHA-256 hashed at the PostgreSQL UTC commit timestamp, creating a tamper-evident chain:

msaHash → paperContentHash → paperSnapshotHash → amendmentHash

From any Paper, you can reconstruct the complete governing terms. The chain is walkable and independently verifiable. This is not blockchain - it is standard cryptographic integrity verification applied to legal documents.

Why This Matters

Enterprise buyers cannot deploy AI agents without governance. Procurement teams require MSAs. Legal teams require liability caps. Compliance teams require audit trails. The SAISA provides all of these in a standardized format that any platform can adopt.

The framework is runtime-agnostic. The same Paper governs an agent running on NVIDIA NemoClaw, Docker, Kubernetes, AWS Lambda, Salesforce Agentforce, or any MCP server. Runtimes execute. The Paper governs.