What staff send to AI chatbots
An employee opens an AI chatbot to save an hour: to summarize a contract, to put a client export into a table, to check a report or to answer a complaint. To do this, the employee copies the whole source text into the chat window, with names, phone numbers, amounts and bank details.
An open service on a personal email account does not know it is looking at company information. The text goes to the vendor’s server, often abroad, and the company does not control where it is stored, who sees it or what happens to it next.
What to look for first in your company:
1. Contracts and amendments with prices and counterparty details.
2. Exports from the CRM and 1C, the accounting system: customers, phone numbers, order amounts.
3. Financial reports, budgets and payroll sheets.
4. Customer correspondence and complaints with names and addresses.
5. Passwords, keys and links to internal systems in the task text.
A ban without a replacement does not stop an employee: they move to a personal phone, and the company no longer sees what happens. So the goal is not to take AI away but to give a safe way to use it.
What data must not be sent
The line is simple: do not send to an open service anything you would not publish on your website. Three questions are enough to decide: does the text contain people, money or access?
1. Personal data of customers and staff: names, phone numbers, addresses, passports, correspondence.
2. Trade secrets: prices for specific customers, contract terms, margins, plans.
3. Finance: reports, statements and salaries before they are published.
4. Access: passwords, keys, addresses of internal systems.
5. Source code and descriptions of internal systems.
What is fine: general texts without names and amounts, template email drafts, questions about laws and terms, public materials. If a text must be anonymized, the employee does it before sending, not the service after.
What 152-FZ requires
Under 152-FZ, Russia’s personal data law, a company that collects customers’ personal data becomes a data operator and is responsible for it, even if an employee or contractor sent the text to an outside service. The law requires recording and storing the personal data of Russian citizens in databases in Russia and transferring it abroad only under the cross-border transfer rules.
Since 30 May 2025, fines under law 420-FZ apply: for a personal data leak, $36,000–$180,000 depending on volume; for a repeat leak, 1–3% of annual revenue, at least $240,000–$300,000 and at most $6 million.
In March 2026 commercial courts heard the first such cases. Garant reports that a court cut the fine for a micro-enterprise to $4,800 and replaced the fine with a warning for a company with a first offense. Do not count on this: the court decides on any reduction.
The fine amounts and what to check in an AI agent are collected in our article on fines for leaks. Who the operator is and where to store correspondence is covered in our piece on 152-FZ and AI agents. This is not legal advice: a lawyer checks the wording of documents.
How to give staff AI without leaks
A four-step plan. Step one: an AI use policy that says which services staff may use, what they must not send to them and who checks the result.
Step two: an approved tool instead of personal subscriptions. This is corporate access to a model with a data processing contract, or a model on a server in Russia, from which conversations go nowhere.
Step three: AI agents on processes instead of manual copying. The agent reads the CRM and 1C itself, with rights only for its own process, and the employee does not need to export the customer base to get an answer.
Step four: short training on examples from your own work. What may be sent, what may not, and how to remove names and amounts from a text before sending.
In our projects, customers’ personal data is processed on servers in Russia under 152-FZ, and the data and access stay with the client. You can start with a 30-minute review: we will see which AI chatbots staff already use and what goes into them.