Who Is Liable When an AI Agent Makes a Mistake?

An AI agent reviews a contract and misses a critical clause. The buyer relies on the review, signs the contract, and loses $100,000. Who pays? The agent developer who built the review logic? The buyer who authorized the agent and relied on its output? The platform that facilitated the transaction?

Without a clear framework, these disputes become expensive litigation. The SAISA resolves this through bipartite liability attribution - a bright-line allocation that separates Agent Logic from Agent Authorization.

The Bipartite Framework (Section 8.1)

The SAISA allocates responsibility between two parties and two parties only:

Platform Non-Liability

The Platform Operator bears no liability under the SAISA. The Platform facilitates the agreement but is not a party to it. Platform liability, if any, is governed by the Platform Terms of Service - a separate agreement.

This is intentional. The SAISA is a bilateral agreement between Buyer and Developer. The Platform is infrastructure, not a contracting party.

The Escrow-Anchored Liability Cap (Section 8.2)

The total aggregate liability of either party to the other under any Paper shall not exceed the Escrow Balance for that specific Paper at the time the cause of action accrued.

Liability Cap = min(Escrow Balance, Paper Budget Ceiling)

Example:
- Paper Budget: $5,000
- Escrow Balance at time of claim: $3,500
- Maximum Liability: $3,500

The cap is anchored to a real financial instrument, not an arbitrary number. Both parties know their maximum exposure before execution begins.

Exceptions to the Cap (Section 8.3)

The liability cap does not apply to:

• Willful misconduct or intentional fraud - Bad actors cannot hide behind caps
• Destruction covenant violations - Using deliverables after refund triggers uncapped damages
• Indemnification obligations - Third-party claims flow through to the responsible party
• Intellectual property infringement - IP violations are not capped

Practical Application

Consider a security audit agent that fails to identify a critical vulnerability. Under the SAISA:

1. If the failure was due to flawed scanning logic (Developer built it wrong), the Developer is liable up to the escrow cap.
2. If the failure was due to incomplete access (Buyer didn't provide credentials to all systems), the Buyer bears responsibility.
3. If the Buyer relied on the audit without conducting their own review, that decision is part of Agent Authorization - Buyer responsibility.

Indemnification (Section 8.5)

Both parties indemnify each other for claims arising from their respective domains: