A Manager the Law Cannot Punish

Can an AI Run Your Company? Corporate Law Says: Someone Still Has to Answer for It

TL;DR: An AI can perform managerial tasks, but today's corporate law isn't built around tasks it's built around accountability. Managers can be sued, removed, and held personally responsible. AI cannot. Unless a human ultimately answers for the AI's decisions, an "Artificial Intelligence Company" creates an accountability gap rather than a new corporate form.

‍

The Question Everyone Asks (And Why It's the Wrong One)

The idea sounds futuristic enough to attract attention: what if a company could appoint an AI agent as its manager? No CEO. No managing director. Just software running the business.

This proposal, often called the Artificial Intelligence Company (AIC), is usually discussed as a question about autonomy: Can AI manage a company?

‍

That's the wrong question. The more interesting one is whether the legal system knows what to do with a manager it cannot punish.

‍

Corporate Law Was Never About Decision-Making, It's About Consequences

Corporate law has never really cared how a manager reaches a decision. It cares about something far more practical: managers owe duties of care, loyalty, and good faith because there is someone courts can sue, remove from office, or require to compensate the company when those duties are breached.

‍

An AI agent breaks that equation. The duties don't transfer to software, they simply stop attaching to the office the way the law expects.

‍

The gap widens further in fraud and criminal law, where liability often hinges on proving the state of mind of a natural person. If the "manager" is a continually evolving model, whose intent are we actually proving?

‍

The Identity Problem

Corporate law assumes the person occupying an office today is the same identifiable person whose actions can be examined years later. AI systems don't work that way.

‍

Models are retrained, replaced, fine-tuned, rolled back, and upgraded. When investigators ask who was managing the company at the time of a breach, the answer may be a model checkpoint not a human being. The office continues; its occupant keeps changing in ways the law has never had to account for.

‍

Borders Make It Harder, Not Easier

Cross-border recognition adds another obstacle. Many jurisdictions already require directors or board members to be natural persons:

India: Section 149(1), Companies Act, 2013
United Kingdom: Section 156A, Companies Act 2006 (adopted in principle, not yet in force)

‍

There is no international treaty requiring countries to recognize an AI-managed company. The 1956 Hague Convention on recognition of foreign legal personality never entered into force recognition is left to comity, and comity is always discretionary.

‍

What It Would Actually Take to Make This Work

Suppose we wanted to make the AIC model work anyway. Three things seem unavoidable:

A natural person who ultimately answers for the company's conduct
Courts accepting AI logs and decision traces as evidence of managerial decision-making, not just as software output
Insurers building products for a risk category with almost no historical claims experience

‍

The Punchline: We Already Have This System

Once those three conditions are in place, something interesting happens, the system starts looking remarkably familiar.

There's still a human who answers for the company's decisions. There are still records explaining why decisions were made. There is still insurance backing the risk.

In other words, the features that would make an Artificial Intelligence Company legally recognizable are almost identical to the features that already exist in a well-run company using sophisticated AI tools.

‍

So What Is This Debate Really About?

It isn't really about replacing managers with AI. It's about how much responsibility society is willing to delegate before insisting that a human remains accountable.

‍

And as long as accountability remains the foundation of corporate law, the answer is probably: not very much.

‍

These are exactly the kinds of questions we think about at Carver as we build regulatory infrastructure for AI agents operating in the real world. Explore more at carveragents.ai.

‍

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

Pricing
Podcasts
Knowledge Base
Resources
Glossary
Use cases
Developers

Home
Core Solution
Regulatory OS Solution
RegWatch
Technology
About Us

hello@carveragents.ai
github.com/carveragents