// wiki · reference · September 2026
AI at work: what an employee may do
Draft article. The rules are as of September 2026; assessing a specific situation depends on the employer's confidentiality regime and the content of the data.
A draft for discussion. The article answers an employee's question “what am I allowed to do”, but does not replace a company policy: where one exists, the policy applies.
The short answer
Using a neural network at work is not in itself prohibited by law and cannot be a ground for dismissal. What is prohibited is what happens around it: passing to a public service information constituting a trade secret, personal data without grounds, clients' confidential documents. In 2026 a court already upheld a dismissal for uploading reports to DeepSeek (No. 02-1545/2026), and the Plenum of the Supreme Court of the RF obliged parties to disclose the use of AI in documents filed with a court. The practical rule: a public service — only for public information; everything work-related — in the company's agreed instruments.
What the court said: the DeepSeek case
In case No. 02-1545/2026 the Babushkinsky District Court of Moscow upheld the dismissal of a sales director for disclosing a trade secret: among the grounds was the upload of reports from the internal protected PowerBI resource to a third-party AI service. The key positions:
- uploading protected information into an AI system is already disclosure: the offence is complete the moment the data leaves the holder's control; proving that third parties saw it is unnecessary (the logic of Constitutional Court Resolution No. 25-P);
- “I used the AI for work purposes” does not save the day — the employee had no production necessity to extract information from protected resources;
- the secrecy regime was formalised: a policy, a signed undertaking, monitoring — which is why the employer won the dispute.
At the same time, lawyers commenting on the case are unanimous: using a neural network is not in itself a ground for dismissal. What matters is which data got into it. The full analysis — in the note “AI and trade secrets”.
If you are a lawyer: disclosing AI to the court
Resolution of the Plenum of the Supreme Court of the RF No. 15 of 21.05.2026 (para. 42): a party presenting facts obtained with the use of AI must inform the court about it. Plus two fines for non-existent practice “selected” by AI (No. А27-7831/2025 — 50,000 ₽ from an organisation; No. А71-11377/2025 — 5,000 ₽ from an individual entrepreneur; the excuse “ConsultantPlus drafted it” was rejected). The practical conclusion: AI for drafts — yes; filing unchecked text — a fine; checking citations against the court register — mandatory.
Checklist: which content into which service
Where it is definitely forbidden (public ChatGPT, DeepSeek, Gemini and the like):
- reports, extracts, dashboards from internal systems;
- client documents and correspondence;
- personal data of employees and clients;
- everything marked with a classification label or covered by an NDA;
- contracts and contract drafts before signing.
Allowed (unless the company policy says otherwise):
- public norms, judicial acts from open court registers;
- your own texts without work data;
- general questions, study tasks, translations of non-secret text;
- drafts of letters containing no identifiable information.
The grey zone is decided by the company: de-identified fragments, synthetic examples, aggregated statistics. The rule here: if in doubt — ask; the company must have a person responsible for this question.
What the company should formalise
- An AI use policy: which instruments are allowed, what data may be passed into them, where to go with a question. Two or three pages, no more.
- A lawful alternative: a closed contour or a corporate subscription with no-training-on-data terms — otherwise the policy will forbid without offering anything in return.
- Data separation: a list of information constituting a secret, need-to-know access, a marking — without a formalised regime there is nothing to protect (Article 10 of Law No. 98-FZ).
- Training: one page of “allowed / not allowed” at onboarding is cheaper than the first dismissal dispute.
Norms: subpara. “v” of para. 6 of part 1 of Article 81 of the Labour Code; Article 10 of Federal Law No. 98-FZ; Federal Law No. 152-FZ; para. 42 of Resolution of the Plenum of the Supreme Court of the RF No. 15 of 21.05.2026.