Russian companies sued over AI-generated work by employees, but courts rarely rule in their favor

Russian companies are increasingly trying to recover damages from employees and contractors who use neural networks as a quick substitute for expertise. But the first disputes show that the mere fact of working with AI does not make a person guilty, and employers have to prove not the technology itself, but specific damage and violation of terms.

Why it came to court

The reason for conflicts is always the same: business expected acceleration but got rework, downtime, and new expenses. In one case, a product manager configured a cloud service following chatbot suggestions and after unsuccessful actions deleted almost all data on the company's servers. Instead of savings, the company lost normal operations for several days and was forced to separately pay for recovery from backups. And this is no longer a single failure, but a symptom of a new practice.

A similar story arises in more "office" tasks. Consultants submit voluminous reports that look solid but don't account for the client's data, answer the wrong questions, and produce machine-like style. In marketing and design, this is compounded by errors in competitor analysis, weak arguments, and the risk of copyright claims if generation copies the recognizable manner of specific artists or photographers. Because of this, the project is formally closed, but in fact has to be rebuilt from scratch.

What courts see

Judicial logic so far is quite pragmatic: it evaluates not whether a person used AI, but what exactly they delivered and what obligations were fixed on paper. If the employment contract, job description, or specific assignment did not prohibit neural networks, and the result was accepted, it becomes much harder for the employer to challenge the payment, recover money, or prove a gross violation.

For the court, the quality of the result is more important than the volume of "manual" labor.

"The customer pays for the result, not for the amount of manual labor

by the contractor."

The problem for companies is that the Labor Code does not yet separately regulate work with AI. Therefore, many come into the process with vague rules that don't describe when neural networks are acceptable, who is responsible for fact-checking, and how damage is recorded. Against this background, an employee can claim that they used an ordinary work tool rather than violated a prohibition that formally didn't exist. This is most often where employers lose.

Where business loses money

Neural networks have already become a mass work tool. According to estimates cited in the material, 45% of Russians use AI at work pointwise, another 36% very actively, while only 15% operate within a corporate perimeter. This explains why the problem quickly went beyond experiments and turned into a management issue: employees delegate to AI texts, data search, design, presentations, and even legal documents, but quality control often remains a formality.

• Errors in code, analytics, and marketing materials have to be rechecked manually
• Legal documents may contain fabricated norms, cases, and references
• Copyright disputes are possible due to generation "in the style of" specific authors
• Parsing poor-quality AI content from colleagues takes hours from the team
• Using public models increases the risk of working outside a protected internal perimeter

The most dangerous area is legal and regulatory texts. An attempt to replace a specialist with a chatbot can result in a document that sounds convincing but relies on non-existent laws or irrelevant precedents. As a result, the business pays twice: first for a quick but weak draft, then for a full check and rewrite. According to experts, about 40% of employees have already encountered poor materials created by colleagues through AI, and parsing one such case takes almost two hours on average.

What this means

The most likely line for business is not AI prohibition but its formalization. Companies will fix rules in contracts and assignments, move employees to internal models, and require mandatory result verification by humans. For workers, this is bad news only in one case: if neural networks are used as a way to submit a raw draft under the guise of finished expertise. Then the dispute will not be about AI, but about work quality and responsibility.