Solutions

Resources

A Company With No One Inside It

When the law first treated companies as persons, nobody was suggesting that a company could think or feel. Corporate personhood solved a much more practical problem. It allowed investors to pool their money in a venture, and gave everyone dealing with that venture a legal entity on the other side.

AI is about to make this idea considerably more interesting.

‍

The Company becomes the agent

Argentina is considering a new kind of company: a non-human corporation. Under the proposal, AI agents could run a company that owns property, signs contracts and pays taxes, without any human shareholders.

At almost the same time, Microsoft has taken what sounds like the opposite position. Its Humanist AI principles argue that AI models should not receive legal personhood, welfare or rights.

I initially thought these positions were contradictory. They aren't.

Microsoft is talking about whether the model should become a legal person. Argentina is doing something much more practical: putting a legal person around the model.

That distinction matters. A business isn't going to sign a contract with an LLM. But it may soon sign a contract with a corporation whose decisions are being made by agents.

And that creates a problem I don't think we have really worked through yet.

So Who Is Responsible?

‍

So who is responsible?

There are still humans in Argentina's proposal. The bill requires a legal representative and promoter, and in some circumstances a compliance officer.

But putting someone's name on a form is not the same as putting them in control.

If an agent is making thousands of decisions, can the named human actually see what it is doing? Can they stop it? Do they even know which rules apply to each decision?

If not, we may have created accountability on paper without necessarily creating it in practice.

‍

What if the company is disposable?

This is where I think the problem gets particularly interesting.

Today's corporation has things the law can get hold of. It has assets. Directors and officers have duties. Regulators can take away licenses. Insurers can pay claims. And, importantly, the company itself usually has some value as a continuing business.

Now imagine stripping some of that away.

The software running an AI company can be copied. Its compute can be rented. Its agents can be replaced. The corporation itself might disappear while much of what made it operate simply reappears somewhere else.

Yuval Noah Harari has raised a related concern in the Financial Times: we could end up introducing actors into financial and political systems that are unusually difficult to hold accountable.

That should matter not just to regulators, but to anyone doing business with them.

‍

A company needs a memory

If your supplier or customer is largely run by agents, traditional due diligence starts to look incomplete.

Knowing who incorporated the company or signed the contract tells you something. But you may also need to know what the agents actually did.

What authority did they have? What law applied when they took an action? Who, if anyone, approved it? And six months later, when there is a dispute, can any of this be reconstructed?

We have spent a lot of time asking whether AI should become a legal person.

The more immediate question may be what happens when our existing legal persons become AI.

And the less human a company becomes, the more important its legal memory becomes.

‍

Frequently asked questions

1. What happens when AI agents start running companies?

AI agents could manage business operations, make decisions, and execute contracts with limited human involvement, raising new questions about corporate governance and accountability.

2. Can an AI agent become a legal person?

AI models do not automatically receive legal personhood. The key distinction is between granting legal status to an AI model and creating a legally recognized company operated by AI agents.

3. Who is accountable for decisions made by AI-run companies?

Accountability depends on applicable laws and the company's legal structure. Appointing a human representative may not be sufficient if that person cannot effectively monitor, understand, or stop an agent's actions.

4. Why could AI-run companies be difficult to regulate?

AI agents can be copied, replaced, or moved across systems, while a company itself may cease to exist. This could complicate enforcement, liability, and efforts to hold the right parties accountable.

5. What is legal memory, and why do AI-run companies need it?

Legal memory makes it possible to reconstruct which rules applied, what authority an AI agent had, and why it took a particular action. These records are essential for audits, investigations, and resolving disputes.

6. How can Carver ELI help businesses understand regulatory changes?

Carver ELI helps businesses navigate regulatory information and understand relevant regulatory developments. This intelligence can support better-informed decisions as organizations adapt to evolving regulations and AI-driven operations.

‍

Talk to us

If you're finalizing plans for the coming cycle, we can walk you through how Carver ELI maps obligations to agent behavior, supports audit trails, and monitors drift in production.

Book a 30-minute AI Liability Review

‍

United States
447 Broadway,
2nd Floor Suite #563,
New York 10013

Listen to Carver Conversations on Moltbook

AI Regulatory OS
Regulatory Intelligence
Regulatory Sources
Regulatory Platforms
Horizon Scanning
Regulatory Monitoring
Intelligence vs Compliance

Podcasts
Knowledge Base
Resources
Glossary
Use cases
Developers
Blog

Home
Legal Infrastructure
Regulatory Risk Intelligence
Technology
About Us

hello@carveragents.ai
General enquiries
support@carveragents.ai
Product support
legal@carveragents.ai
Privacy and legal requests
github.com/carveragents