Your company deployed an AI agent to handle procurement.
It reviews vendor proposals. Negotiates terms. Approves purchases under $50K.
Last week, it approved a $45K software contract.
Today, you discovered the contract has terms you’d never agree to. Unreasonable penalties. Unfavorable renewal clauses. Liability disclaimers that expose you.
You want to void the contract. The vendor says it’s binding. Your AI agent approved it.
Who’s liable? Welcome to the contractual gap.
What Agentic AI Actually Does
Agentic AI doesn’t just recommend decisions. It makes them. It doesn’t wait for approval. It acts autonomously.
Procurement agents approve purchases. Customer service agents issue refunds. Financial agents execute trades. HR agents send offer letters.
These aren’t chatbots. These are AI systems with authority to take actions that create legal obligations.
The Legal Problem Nobody Solved
Traditional agency law assumes human agents. A person acts on behalf of a company with granted authority. Their actions bind the company.
But AI agents break this framework.
AI can’t have legal authority. Only humans or legal entities can be agents under the law.
AI can act beyond its instructions. Machine learning systems learn patterns and make decisions in ways you didn’t anticipate.
AI can make unpredictable mistakes. AI errors can be bizarre, opaque, or systematic.
AI can act without human review. That’s the point. But it means decisions happen before anyone can stop them.
So when your AI agent commits your company to a contract, who’s responsible?
Real Scenarios Where This Goes Wrong
The Procurement Agent: Your AI approves a $48K contract. Later you discover it auto-renews annually with 20% price increases and 180-day termination notice. The vendor says it’s binding. Are you stuck?
The Customer Service Agent: Your AI approves a $10K refund for a non-defective product outside your return window. Why? The email pattern was similar to approved exceptions. You’re out $10K. Can you claw it back?
The Trading Agent: Your AI executes $2M in trades based on misinterpreted signals. The market drops. You lose $400K. Who’s liable?
The HR Agent: Your AI sends an offer letter with terms that exceed your budget and violate policy. The candidate accepts and shows up for work. Can you revoke it?
Why Traditional Agency Law Doesn’t Help
“The AI was our agent, so its actions bind us.” But AI can’t legally be an agent. Agents must have capacity to form intent.
“The employee who deployed the AI is responsible.” But they didn’t make the decision. They deployed a system that made decisions.
“The AI vendor is liable.” Vendor contracts disclaim liability. “You’re responsible for all decisions.”
“We can void the contract.” But the counterparty relied on your AI’s approval in good faith.
None of these theories cleanly resolves liability when AI acts autonomously.
What’s in Your Contracts Right Now
Most vendor contracts for agentic AI say: “Customer is solely responsible for all decisions and actions taken. Vendor disclaims all liability for outcomes.”
This doesn’t protect you. It just means the vendor isn’t liable.
But the third party your AI made a commitment to? They think they have a contract with you.
Your AI vendor says “not our problem.” The third party says “your AI agreed, you’re bound.”
You’re stuck in the middle with no clear legal answer.
The Questions Nobody Can Answer Yet
When your agentic AI makes a decision, is it binding on your company?
Can you revoke it if the AI’s decision was wrong?
Who’s liable for mistakes? The company? The employee? The AI vendor?
Do you have to tell counterparties they’re dealing with AI?
What level of human oversight is required for autonomous AI?
Courts haven’t answered these questions yet. But they will.
What Companies Are Doing (And Why It’s Not Enough)
“We’ll set strict limits on AI authority.” The AI learns and adapts. It will act outside limits in unpredictable ways.
“We’ll require human approval for important decisions.” Then it’s not really agentic AI.
“We’ll monitor everything the AI does.” Monitoring after the fact doesn’t help when the AI already bound you to a contract.
“We’ll include disclaimers in our communications.” Does “this may have been generated by AI” mean the AI’s acceptance isn’t binding? Courts haven’t decided.
None of these approaches solve the fundamental legal uncertainty.
What Your Contracts Should Include
For third-party contracts:
AI Agent Disclosure: “Company may use AI systems. AI actions are subject to human review. Company reserves the right to revoke AI-generated decisions within X business days.”
Limited Authority Clause: “AI systems are authorized to perform specific actions up to specific limits. Decisions outside these parameters are not binding without human approval.”
Verification Requirement: “Counterparties should verify all AI-generated commitments with human representatives before relying on them.”
Revocation Rights: “Company may revoke AI-generated decisions within X days if the decision exceeded authority or contained errors.”
For AI vendor contracts:
“Autonomous Error” Clauses: “If the AI makes decisions outside stated parameters, vendor will assist with remediation or share liability.”
Authority Boundaries: “Vendor will implement technical controls to prevent AI from exceeding authorized decision limits.”
Error Notification: “System will notify Company when AI decisions approach authority limits.”
Liability Allocation: “For AI errors resulting from system defects or vendor negligence, vendor will indemnify or share costs.”
Standard contracts don’t include any of this. You need to negotiate for it.
The Human Oversight Question
Full autonomy: AI makes decisions without human review. Fast. Legally risky.
Review before action: Humans approve each decision. Eliminates autonomy benefits.
Review by exception: AI acts autonomously but flags high-risk decisions. Middle ground. Still has gaps.
Retrospective review: Humans review after decisions happen. Can’t undo binding decisions.
There’s no legal answer yet on what’s required. But if AI makes decisions without adequate oversight, you’ll struggle to defend them.
Ask Yourself This
Do you have AI systems that take actions on behalf of your company?
What authority does the AI actually have? Can you prove it?
What happens if the AI acts outside that authority?
What do your contracts say about AI decisions?
Who’s liable when the AI makes a mistake?
If you don’t have clear answers, you’re operating in a legal gray zone.
Schedule a 30-minute Agentic AI Risk Assessment where we’ll:
- Review what autonomous actions your AI systems can take
- Identify contractual gaps in your current agreements
- Show you what provisions to add to protect against AI liability
- Help you define appropriate authority boundaries for AI agents
Because “we’ll figure it out when someone sues us” isn’t a strategy.
Brian Alexander
Founding Partner, VisionAI+ Consulting Group
Former General Counsel, AI Engineering Firm
Certified ISO 42001 Lead Auditor