The liability for a mistake of artificial intelligence lies not with the AI itself but with the people and organizations that created and used it. This page describes Russian law. An AI agent is not a party to civil-law relations: under the Civil Code of the Russian Federation, the parties are individuals, legal entities and public-law entities. The company that used the agent answers to a client or counterparty for the agent’s action, and its chief executive answers to the company itself for losses, if he or she acted unreasonably or in bad faith (Article 53.1 of the Civil Code).
Quotations from Russian legal acts below are our translation; the official texts are in Russian.
In brief
- AI is not a subject of law. The parties to civil-law relations are individuals and legal entities, as well as the Russian Federation, its constituent entities and municipalities (Civil Code, Article 2, point 1). The AI law, No. 243-FZ, introduces no special liability regime (Article 11).
- The company answers to the counterparty. In business activity, regardless of fault, unless it proves force majeure (Civil Code, Article 401, point 3).
- The chief executive answers to the company for losses, if it is proven that he or she acted in bad faith or unreasonably (Civil Code, Article 53.1).
- In bankruptcy: subsidiary liability of the controlling person for the company’s debts (Article 61.11 of the Insolvency Law).
- The model’s developer answers under its contract with the company and under the general rules; the law sets no separate duty for it to answer for every action of someone else’s agent.
Can an AI agent itself be liable for harm?
No. Under Article 2, point 1 of the Civil Code, the parties to civil-law relations are individuals and legal entities; the Russian Federation, its constituent entities and municipalities may also take part in them. A program is none of these: it has no property from which harm could be compensated, and it cannot be held liable.
Federal Law No. 243-FZ of 26 July 2026 “On Supporting the Development of Artificial Intelligence Technologies” does not change this. Its Article 11 states that participants in these relations are liable in accordance with the legislation of the Russian Federation, that is, under the general rules. The law is described in detail on the page about Law No. 243-FZ.
The EU AI Act also distributes obligations between people and organizations, the providers of AI systems and those who use them (deployers), and not the systems themselves. More on the page about the AI Act.
Who is liable if an agent made a payment or deleted data?
The company that used the agent. There are two kinds of liability.
To a counterparty under a contract. If an agent disrupted the performance of a contract, by shipping the wrong goods, sending money to the wrong place or deleting a client’s data, the company is liable as the party that breached the obligation. For businesses the rule is strict: a person who breaches an obligation in the course of business activity is liable unless it proves that performance became impossible because of force majeure, that is, “extraordinary circumstances that could not be prevented under the given conditions” (Civil Code, Article 401, point 3). The article expressly names circumstances that do not count as such: breaches by counterparties, the absence of the necessary goods on the market, and the debtor’s lack of money. A mistake of the company’s own tool is hard to present as an extraordinary and unpreventable circumstance; the final qualification belongs to the court.
To third parties outside a contract. Harm to property or to a person is compensated in full by the person who caused it; that person is released if it proves the harm was not caused through its fault (Civil Code, Article 1064).
In both cases, the phrase “the AI did it” does not release the company by itself: the agent is the company’s tool, not an independent party.
Can a company back out of a transaction that an agent made beyond its rules?
Do not count on it. For people, the Civil Code resolves a similar case this way: a transaction made by a representative without authority or in excess of it is not void; it is deemed concluded in the name and interests of the person who made it, unless the principal later approves it (Civil Code, Article 183, point 1). An agent, however, is neither a representative nor a person, so this article does not apply to it directly. How a court would qualify a transaction that an agent made beyond the rules set for it is still an open question.
The practical conclusion: the rules for what an agent may do must work before the action is executed. Undoing it afterwards is expensive and not always possible.
When is the chief executive personally liable?
The chief executive answers to the company, not to its clients. Under Article 53.1 of the Civil Code, a person authorized to act on behalf of a legal entity must, at the request of the legal entity or its members, compensate losses caused through his or her fault. Liability arises if it is proven that the executive acted in bad faith or unreasonably, including when his or her actions did not correspond to the ordinary conditions of civil commerce or the ordinary business risk (Article 53.1, point 1). Members of collegial bodies are liable in the same way, except those who voted against the decision or in good faith did not take part in the vote (point 2), and so is a person who in fact determines the company’s actions (point 3).
Resolution No. 62 of the Plenum of the Supreme Arbitration Court of the Russian Federation of 30 July 2013 explains what counts as unreasonable. Unreasonableness is considered proven, in particular, when the executive (point 3):
- made a decision without taking into account information known to him or her that mattered in the situation;
- before making the decision, did not take the steps to obtain the necessary and sufficient information that are usual in business practice in similar circumstances;
- made a transaction without following the internal procedures usually required or adopted in that legal entity.
Separately (point 5): the executive is liable for losses when he or she chose, or controlled the actions of, representatives, counterparties and employees in bad faith or unreasonably, and when the management system of the legal entity was improperly organized.
For deploying AI agents this means the following. If an agent was given rights to payments or data without a risk assessment, without rules and without control, and it caused losses, the question to the chief executive will be whether a management system for this tool was organized at all. More in the book Proven Action (in Russian).
How do losses under Article 53.1 differ from subsidiary liability?
| Losses (Civil Code, Article 53.1) | Subsidiary liability (Insolvency Law, Article 61.11) | |
|---|---|---|
| Who recovers | The company or its members | Creditors, through bankruptcy proceedings |
| When | At any time while the company operates | Only in a bankruptcy case |
| To whom the person answers | To the company | For the company’s obligations to creditors |
| What is proven | Losses, bad faith or unreasonableness, and the link between them | That the claims cannot be paid in full because of the controlling person’s actions; presumptions apply |
| Who is liable | The executive, members of collegial bodies, the person in actual control | The person controlling the debtor |
One of the presumptions of Article 61.11 is directly tied to record-keeping: the controlling person’s fault is presumed if, by the time of bankruptcy, mandatory accounting or reporting documents are missing or distorted and this makes the proceedings substantially harder (point 2, subpoint 2). From 27 July 2027, point 2 of Article 61.11 is amended (Federal Law No. 253-FZ of 26 July 2026).
What counts as proof of reasonableness and good faith?
The Plenum of the Supreme Arbitration Court (point 1 of Resolution No. 62) proceeds from the premise that negative consequences by themselves do not prove that the executive acted in bad faith or unreasonably, and that the court does not review the economic expediency of his or her decisions. The claimant must prove the breach; the executive may explain the causes of the losses and present evidence. If the executive refuses to explain or the explanations are clearly incomplete, the burden of proof may shift to him or her.
For a decision to deploy an AI agent, the evidence of reasonableness is documents that existed before the incident:
- a risk assessment: what actions the agent can perform and what harm they can do;
- approved rules for what the agent may and may not do: amounts, recipients, data, deletion;
- internal procedures: who approved the agent’s rights and who is responsible for control;
- a log of the agent’s actions that cannot be rewritten after the fact and that a third party can re-verify;
- incident reviews: what happened and what was changed afterwards.
This is exactly the set that the IS 3.0 architecture describes: a rule, a check of every action before it runs, and a log that can be re-verified.
Is the model’s developer liable?
To the company, under its contract with it: a licence, a services contract, a development contract. The limits of that liability are set by the contract itself and are worth checking before deployment. To injured parties, under the general rules on compensation for harm, if the harm was caused by the developer’s actions. Law No. 243-FZ gives the developer of a sovereign or national model the right to insure liability for harm caused when the model is used (Article 7, point 5), but does not make it liable for every action of an agent built on its model.
Can liability for AI mistakes be insured?
The market is only taking shape (in detail, on the page about cyber insurance). In the United States, since January 2026 there have been optional ISO (Verisk) endorsements that exclude generative AI from commercial general liability (CGL) policies; insurers attach them at their discretion. At the same time, separate policies are appearing: in January 2026, Armilla announced an AI liability policy with a limit of up to $25 million per organization, backed by Lloyd’s underwriters, with an “AI agent failures” category. The AIUC-1 standard on which such policies are built requires logging the full execution chains of agentic processes, including human approvals, and protecting the logs against tampering.
How does this work in other countries?
- European Union. The AI Act places obligations on the providers of AI systems and on those who use them; fines reach €35 million or 7% of worldwide turnover for prohibited practices (Article 99). More on the page about the AI Act.
- China. The “Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents” (CAC, NDRC, MIIT, 8 May 2026) are a policy document, not a law. Their point 6 requires separating the user’s own decisions, decisions made on the user’s instruction and the agent’s independent decisions, keeping the final decision with the user, and states that an agent may not act beyond its authority. More on the page about China’s rules for AI agents.
What to do
- Assign a responsible person to every agent: someone who approved its rights and is responsible for control.
- Assess the risks before deployment and keep the assessment: it will be needed if the chief executive has to prove reasonableness.
- Write down the agent’s rules, what is allowed, what is not, with what limits, and check every action against them before it runs, not after.
- Keep a log of the agent’s actions that cannot be rewritten after the fact.
- Check the contracts with model providers and with clients: who answers for an agent’s mistake, and within what limits.
- Check the insurance policy for AI-related exclusions.
Frequently asked questions
Who is liable for the mistakes of AI?
Not the AI itself: it is not a party to civil-law relations (Civil Code, Article 2, point 1). The company that used the system answers to clients and counterparties, in business activity regardless of fault, unless it proves force majeure (Civil Code, Article 401, point 3). The chief executive answers to the company for losses, if he or she acted in bad faith or unreasonably (Civil Code, Article 53.1).
Can AI be a subject of law in Russia?
No. The parties to civil-law relations are individuals and legal entities, as well as the Russian Federation, its constituent entities and municipalities (Civil Code, Article 2). The AI law, No. 243-FZ, does not give AI legal personality and refers to general law (Article 11).
Is the chief executive liable for the actions of an AI agent?
To the company, yes, if it is proven that he or she acted unreasonably or in bad faith (Civil Code, Article 53.1). According to Resolution No. 62 of the Plenum of the Supreme Arbitration Court, unreasonableness is proven, in particular, if a decision was made without the necessary information or without following internal procedures, and also when the management system was improperly organized.
Can a transaction that an AI agent made by mistake be cancelled?
Do not count on it. Article 183 of the Civil Code, on a transaction made without authority, speaks of human representatives, and such a transaction is not void. How a court would qualify an agent’s transaction made beyond its rules is an open question. It is more reliable to check the agent’s action against the rules before it runs.
How can a company prove that it acted reasonably when deploying AI?
With documents that existed before the incident: a risk assessment, approved rules for the agent, internal procedures for approving its rights, and a log of the agent’s actions that cannot be rewritten after the fact.
Sources
- Civil Code of the Russian Federation (Part One), as amended on 10 June 2026: Articles 2, 53.1, 183, 401 (in Russian), ConsultantPlus: https://www.consultant.ru/document/cons_doc_LAW_5142/
- Civil Code of the Russian Federation (Part Two), as amended on 24 June 2025 with changes of 9 June 2026: Article 1064 (in Russian), ConsultantPlus: https://www.consultant.ru/document/cons_doc_LAW_9027/
- Federal Law No. 127-FZ of 26 October 2002 “On Insolvency (Bankruptcy)”, as amended on 4 August 2026: Article 61.11 (in Russian), ConsultantPlus: https://www.consultant.ru/document/cons_doc_LAW_39331/
- Resolution No. 62 of the Plenum of the Supreme Arbitration Court of the Russian Federation of 30 July 2013 “On Certain Issues of Compensation for Losses by Persons Who Are Members of the Bodies of a Legal Entity” (in Russian), website of the Supreme Court of the Russian Federation: https://www.vsrf.ru/documents/all/17926/
- Federal Law No. 243-FZ of 26 July 2026 (in Russian), ConsultantPlus: https://www.consultant.ru/document/cons_doc_LAW_540336/
- Armilla, announcement of 21 January 2026: https://www.armilla.ai/resources/armilla-ai-raises-lloyds-backed-coverage-to-25m-as-traditional-insurers-retreat-from-ai-risk ; the AIUC-1 standard: https://standard.aiuc-1.com
- CAC, “Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents”, 8 May 2026 (in Chinese): https://www.cac.gov.cn/2026-05/08/c_1779979789523320.htm
Checked: 16 October 2026. The legal conclusion in a specific situation belongs to the organization’s lawyer.
Related pages
- Russia’s law on artificial intelligence: Federal Law No. 243-FZ
- EU AI Act: dates and requirements after Regulation 2026/1744
- AI agent incidents: deleted databases, files and services
- Cyber insurance and the risks of AI agents
- The book “Proven Action. Power and Control over AI Agents: Strategy 2027–2035” (in Russian)