The habit of uploading work files to a chat with a neural network seems harmless until the employer discovers a leak of trade secrets. This is exactly what happened to a top manager of a Moscow engineering company, which recovered damages from a former employee through the court.

Personal email in BCC and documents in DeepSeek

The plaintiff held the position of sales director for less than six months with a salary exceeding 800 thousand rubles per month. After being dismissed at the employer's initiative, she demanded through the court to cancel the order, insisting on terminating the contract by mutual agreement with a compensation payment of 5 million rubles.

The company did not recognize the claim, stating repeated disclosure of trade secrets. According to the employer, the employee forwarded official documents via corporate email, placing her personal inbox in BCC. She uploaded a number of documents from the company's internal secure resource to the Chinese neural network DeepSeek, creating a risk of information interception.

The company's arguments also mention that after the disclosure, the counterparty stopped responding during negotiations. Separately, the employer pointed out the systematic failure to meet sales plans. According to the company's version, compensation payments were only provided for in cases of staff reduction or when the position was no longer needed.

The court concluded that the plaintiff had no operational need to extract information from the employer's local secure storage devices. It was not permissible to forward it to an external uncontrolled address or upload it to the resource of the third-party service DeepSeek. The court classified sending confidential information to domains not approved by management, including personal email, as a gross violation of labor duties.

During the case review, the defendant offered a settlement agreement, a change in the wording of the dismissal, and a voluntary payment of over 400 thousand rubles. The plaintiff refused. The court found no grounds to deem the order unlawful and rejected the claims.

Why a neural network is a bad place for work files

This story coincided in time with another news item: last week, several media outlets reported that chats of Russian DeepSeek users appeared in Google search results and became publicly accessible. The cause could be a poorly configured access rights system by the developers.

Here is how to avoid problems when communicating with AI on work matters:

  • clarify in local regulations which services are permitted for working with internal documents;
  • use corporate subscriptions instead of free web versions;
  • anonymize materials before uploading, removing counterparty names, amounts, and details;
  • remember that outgoing corporate email traffic is logged, and BCC is not truly hidden.

Expert opinion: This case is a clear signal for the market. In the era of widespread AI adoption, companies must strictly regulate employees' use of external neural networks, and workers themselves should remember: the digital footprint is not erased, and employer trust is an asset easily lost with one careless click.