⏮First Firing for AI: Court Sides with Employer⏭

On July 27, it became known that a Moscow court upheld the dismissal of a top manager who uploaded work documents to DeepSeek. The sales director with a salary of over 800,000 rubles had worked at the engineering company for less than six months and lost her position under the article for disclosure of commercial secrets. The court denied all her claims, including compensation of 5 million rubles.

🔡Details of this story
There were two leak channels: documents went to a personal email in hidden copy and were simultaneously uploaded from a secure system to a chatbot. Budget plans and employee salaries with names leaked. A separate episode involves the transfer of information to a supplier, which caused the company not to sign a contract. The dismissed woman believes that they simply decided to get rid of her and plans to appeal the decision.

➡️ But behind this particular case lies a problem that concerns almost all of us.

Russia has no frontier-level models of its own. So if you understand that today you can't get far without language models, the choice comes down to two directions: America or China.

Just a month ago, US leadership was indisputable. Today, Chinese models have come so close that the difference is almost imperceptible. With one caveat: the Chinese are great at catching up, but breakthrough things are still born in American laboratories, and then they are picked up.

In terms of money, the gap is more noticeable. A near-top Anthropic model consumes about three times more budget than a comparable Chinese counterpart. Although even here it is possible that we are witnessing a market war with corresponding pricing.

🔡 The main question is different
In both cases, data goes abroad, and the only difference is where it is better preserved. In our humble opinion, the regulation of confidential information in the US is more developed: there they come for such things and ruin you with substantial sums. In China, data retention looks like a much more mundane practice.

As for the choice of the dismissed woman, the LLM provider is debatable, of course. Some sources hinted that she didn't even access it directly, but through intermediaries, meaning the data passed through another link. But even more interesting is another question: how did the employer find out about this at all? Did she herself show that she knows how to work with a neural network, or was the reason sought deliberately? There are plenty of pitfalls in this story.

The moral is not that AI at work is evil. It's that there are no rules yet, and everyone defines them for themselves. The reasonable minimum is known: anonymize confidential information, and don't take critical things outside at all.

🌐 Website | 🎰LinkedIn