Quick answer
Not as a board member. Estonian law requires every board member to be a living person, so an AI board member cannot be appointed. AI can still do much of the work: bookkeeping, payments within limits, routine contracts, even tasks once done by staff. The company is bound by what its AI agent does, and the human director answers for how the agent was set up and supervised.
Agentic AI now writes to suppliers, reconciles bank statements and starts payments. Tech leaders talk about the “one-person billion-dollar company”. For an Estonian company that already runs online, often managed by an e-resident founder abroad, the next step looks obvious: let an autonomous agent run it.
So far, attempts like this have made headlines rather than law. When a Hong Kong venture firm announced an algorithm as a board member, the would-be AI director could not be appointed under local law and was left with observer status.
Can AI Be a Board Member? What Estonian Company Law Says
The Commercial Code (Äriseadustik) settles the question in one sentence. Under § 180(2), a management board member of a private limited company (OÜ) need not be a shareholder but must be a natural person with active legal capacity.
An AI is not a person in law, so an AI board member cannot be appointed, and an AI cannot even hold a power of attorney. Nor can it be slipped in behind a company, because a legal person may not sit on an Estonian management board at all. Ownership follows the same logic:
| Role | Can an AI fill it? | Who it must be |
|---|---|---|
| Management board member | No | A natural person with active legal capacity |
| Shareholder | No | A natural or legal person |
| Beneficial owner | No | Always a natural person; if no one qualifies through ownership or control, a board member is registered instead |
Our guide to the role of a management board member covers the duties of the office in detail.
Can AI Run a Company on Its Own? The Zero-Person Company Test
Not yet, and not legally. KPMG and the University of Amsterdam are testing a “zero-person company”: a web shop led by an AI CEO and overseen by a human supervisory board. When agents were given full executive roles such as CFO, they drifted off task, invented facts and ignored instructions. Things began to work only once processes were broken into small tasks for narrow agents, with people setting the boundaries.
Estonian law points the same way. AI can run a business day to day, but not without people. An Estonian OÜ may have no employees, yet its smallest legal core is one person acting as shareholder, director and beneficial owner at once. Replacing staff with “AI employees” is straightforward, as software needs no employment contract and carries no payroll tax. Two things do not change, though:
- The tax authority and banks look for people. When a company applies for a VAT number or a bank account, they ask who actually runs the business and from where. A company whose only “staff” is software should be ready for those questions; see why VAT numbers get refused.
- AI does not change where the company is managed. For tax purposes, what counts is where the people who make the key decisions work. A founder abroad who runs an Estonian company through AI agents may still create a taxable presence in the country where they live, as our guide to running an Estonian company from abroad explains.
What AI Agents Can Automate in an Estonian Company
The law draws a clear line. Acts reserved for a person stay with a person, and the work around them can be handed to AI.
| Area | What an AI agent can do | What a person must do |
|---|---|---|
| Bookkeeping | Categorise transactions, match receipts, draft entries | Answer for the accounts; a board member signs the annual report |
| Invoices and payments | Read invoices, prepare payment batches, pay within set limits | Grant bank access rights and approve payments above the limit |
| Contracts | Draft, compare and negotiate within agreed rules; accept standard online terms | Sign significant contracts with a qualified e-signature |
| Taxes | Prepare VAT and payroll returns | File them in e-MTA as a board member or an authorised person |
| Staff | Draft schedules, assign routine tasks, sort applications | Decide on hiring, pay, evaluations and dismissals |
Whatever tools you use, one rule holds: never give an agent your PIN codes. A qualified e-signature has the same legal effect as a handwritten one, so anything signed with your codes counts as signed by you.
Using AI to Manage Employees
Managing people is the most sensitive use of AI in business. The EU AI Act classes systems used to recruit, decide on promotion or dismissal, allocate tasks based on personal traits or monitor performance as high-risk. An employer must inform staff before using such a system. AI that infers employees’ emotions at work is banned outright, except for medical or safety reasons. The GDPR also gives people the right not to be subject to significant decisions made solely by automated means. AI can organise the work, but a person should make the decisions about people. Our article on AI Act supervision in Estonia explains who enforces these rules.
AI Agent Liability: Who Pays for an Agent’s Contracts, Payments and Mistakes?
In law, an AI agent is a tool, much like accounting software or a web shop’s checkout. When it gets something wrong, three parties may be asked to pay: the company, its director and the AI provider.
The Company Is Bound by Its AI Agent’s Contracts and Payments
A declaration sent through an automated system counts as the declaration of the business that runs it. If your agent accepts an offer or pays an invoice within the authority you gave it, the company has acted. A contract can be undone for a mistake only in narrow cases.
The company also answers when its agent breaks the rules. In a test run by Andon Labs, an AI shopkeeper refused refunds it had promised and proposed price-fixing to competitors. In a real business, the consumer claims and competition fines would have landed on the company, not on the software.
The Director Answers for How the Agent Was Set Up
Under § 187 of the Commercial Code, a board member must act with the diligence of a prudent businessperson. Board members who damage the company by breaching their duties are jointly liable for it, and a board member escapes liability only by proving they acted with that diligence. “The agent did it” is no defence. A court would ask whether the tool was a reasonable choice, whether its powers were limited and whether anyone checked its work.
Unpaid taxes can land on the director as well. Where the debt arose because the director breached their duties intentionally or through gross negligence, the Tax and Customs Board can recover it from them personally under § 40 of the Taxation Act.
The AI Provider Rarely Pays
AI providers usually cap their liability tightly in their terms of service, and the revised EU Product Liability Directive compensates only natural persons, not a company’s business losses.
The table below shows who carries the loss in three typical cases.
| What goes wrong | Who bears it | Director’s risk | AI provider’s risk |
|---|---|---|---|
| Agent accepts unfavourable terms | The company is bound by the contract | If the agent had no limits or oversight | Only as its terms allow |
| Agent pays a fake invoice | The company; an authorised payment is hard to recover | If there were no payment controls | Rarely any |
| Agent files a wrong tax return or misses a payment | The company owes the tax, interest and any penalty | Personal liability for the tax debt in serious cases | Only as its terms allow |
How to Delegate to an AI Agent Safely: Five Safeguards for the Board
These safeguards turn an open risk into a managed one, and each doubles as evidence of due diligence:
- Write the mandate down. A board resolution stating what the agent may do, and up to what amount.
- Limit the money. A separate account, a virtual card or payment limits, with human approval above a set threshold.
- Keep the logs. What the agent was asked, what it did and what it sent.
- Route exceptions to a person. New counterparties, unusual terms, large amounts and any decision about people.
- Check the provider and your insurance. Liability caps, data terms, and whether your directors’ liability or cyber policy covers losses caused by automated systems.
Will AI Ever Get Legal Personality and a Seat on the Board?
Not in the foreseeable future, so an AI board member will remain an idea rather than a register entry. The European Commission’s expert group on liability concluded that AI needs no legal personality, because any harm it causes can be attributed to existing people or organisations. The experts behind Estonia’s national AI strategy took the same view. The AI Act regulates artificial intelligence as a product, with duties for those who build and use it, as our guide to the EU AI Act for companies explains.
Even Wyoming, the US state that lets a DAO company be “algorithmically managed” by smart contracts, requires those contracts to remain upgradable and the company to keep a registered agent, so people stay in control. AI will take on more and more of the work, but a named person will still answer for it.
Setting Up an AI-First Company in Estonia With Legal Support
Estonia is one of the easiest places in Europe to run a fully digital company, and an AI-first business model suits it well, provided the human side is set up correctly from the start. Eesti Firma’s lawyers can review how you plan to use AI agents in your company. We can draft the board resolution that defines an agent’s mandate, check provider terms, and assess what the AI Act and the GDPR require in your case. If you are just starting out, we can also register your Estonian company with a management set-up that fits the way you intend to work.
Frequently Asked Questions
Yes, as an internal tool. An AI can analyse proposals and present recommendations at board meetings, but it is not entered in the Commercial Register and has no vote in law. Every decision remains the decision of the human board members.
Not safely. The director keeps the full duties and liability of the office, because Estonian law does not recognise a purely formal board member. Approving decisions without understanding them is itself a breach of the duty of care.
Only in narrow cases: when the error is material and the other party caused it, knew or should have known about it (as with an obviously wrong price), or made the same mistake. If the agent simply agreed to poor terms, the company is bound by them.
No. Banks open accounts through the company’s board member and check its beneficial owners, all of whom are people. Once the account exists, an authorised person can give an agent limited access, for example through the bank’s interface or a card with a spending limit.
AI can handle much of the routine bookkeeping, but the board remains responsible for the accounts, and a board member must sign the annual report. Many companies pair AI tools with an accountant who checks the result and handles the filings.
Often, yes. Under the EU AI Act, AI systems that interact directly with people must let them know they are dealing with AI, unless that is obvious. Being open about it also helps avoid later disputes over what was agreed.
