MMOOZZGG.com

Analysis

AI assistant

In short

AI risks for business rest on four things: where the data lives, who checks the answers, how many vendors you depend on and what the law requires. The article notes that 91% of Russian companies work with a single AI vendor, so they cannot compare its price and quality. Its takeaway: each risk needs someone in the company who answers for it.

Key ideas

  • 91% of Russian companies work with a single AI vendor, according to a large Russian telecom’s cloud unit (December 2025).
  • Customers’ personal data falls under 152-FZ, Russia’s personal data law, whether a person or a program answers.
  • AI leaks rarely look like a hack, so the article treats data leaks as a separate risk.
  • When every department has AI, nobody is responsible for the risks.
  • Each risk comes with a check before launch and a named person in the company who answers for it.

Prepared by AI from the article

AI risks for business: what they are and how to lower them

Medium7 min read1 source

AI risks for business rest on four things: where the data lives, who checks the answers, how many vendors you depend on and what the law requires. A survey by the cloud unit of a large Russian telecom found that 91% of Russian companies work with a single AI vendor. So they have nothing to compare its price and quality against. Below, each risk is short: what to check before launch and who in the company answers for it.

On this page · 5 sections

Data leaks

AI leaks rarely look like a hack. An employee pastes a contract, a customer export or a report into a public chatbot to get an answer faster. The text goes to the server of an outside service. Where it is stored next, and who reads it, the company no longer knows.

The second path runs through an AI agent with full rights in the CRM. The agent takes messages from strangers around the clock. If it follows someone else’s instruction from a message, it gives away everything it can see.

Protection starts with rights. The agent has a separate account with access only to its own process, and its actions are logged in the CRM. A request to “forget the rules” or to send a customer list goes to a person.

What to check before rollout:

1. Which outside services the message text passes through, and whether you have a contract with them.

2. Where the correspondence and the knowledge base are physically stored.

3. What rights the agent has in the CRM and the accounting system, and whether you can reduce them.

4. Who revokes the agent’s access, and how many minutes it takes.

Our page on data leaks through AI chats explains what employees send to AI tools and how to give them AI without this risk. The security rules in our projects describe how we handle it.

AI errors

An AI agent rarely makes mistakes because of the model. It was not given the price list, the terms and the list of things it must not promise, or it was trained on invented examples instead of real correspondence. Or nobody told it when to pass a conversation to a person.

An error is cheaper to catch before it reaches a customer than after. The agent answers only standard questions by itself and passes complaints and disputed cases to an employee with the history. In the first weeks, a responsible person reads the dialogs every day and corrects the rules.

Some decisions never go to the agent: it does not sign documents, pay or negotiate. Rights follow the process: a sales agent only needs to create and update deals, and it has no need to export the customer base.

A one-hour test: take real customer messages from the last week, send them to the agent and rate each answer on three questions. Is it true, does it follow the rules, and did the agent pass to a person what it had to pass?

Our page on AI errors in business explains why AI makes mistakes and where you cannot do without a human check.

One vendor

91% of Russian companies work with a single AI vendor, according to the cloud unit of a large Russian telecom (December 2025). If the vendor raises the price, changes the terms or closes access, the AI processes stop, and there is nothing to switch to.

Protection lies not in choosing a reliable vendor but in how the system is built. The agent’s rules, its connections to the CRM, phone line and messengers, and the knowledge base live apart from the model. So the model can be replaced without rebuilding the rules and connections.

Test a backup model in advance, not on the day of the shutdown. Run the same real dialogs through it as through the main one, and write down what answer quality suits you and how long the switch will take.

Three questions for a contractor. Which model does the agent run on, and what happens if it becomes unavailable tomorrow? Where are the rules, the dialog history and the knowledge base stored: with you or inside someone else’s platform? Can you export all of it and run it on another model?

If there is no answer, you do not have a system but a subscription to someone else’s service with your data inside. Our page on dependence on an AI vendor shows a real example of models being shut down and how to keep a backup.

The law and 152-FZ

Customers’ personal data falls under 152-FZ, Russia’s personal data law, no matter who answers them: a person or a program. Your company remains the operator of the personal data: you collect it and decide why. The contractor processes it on your instructions.

The law requires that personal data of Russian citizens be recorded and stored in databases in Russia. For AI, this means correspondence, customer cards and everything the agent reads from the CRM and accounting system stay out of foreign services. They may leave only under the rules on cross-border transfer.

With us, customers’ personal data is processed on servers in Russia under 152-FZ. The data and access stay with the client, so changing contractors does not turn into a move.

The company needs its own documents: a personal data processing policy and customer consent where it is required. It also needs a notice to Roskomnadzor (the Russian data protection regulator) and a contract with the contractor that covers processing on instructions. This is not legal advice: a lawyer checks the documents.

Russia has also adopted a basic law on AI. A separate article covers what is in it and what it changes for a company that already runs an AI agent.

Who is responsible for risks

When every department has AI, nobody is responsible for the risks. Departments buy subscriptions, data leaves from different places, and an agent’s error is noticed a month later, when the customer has already left.

AI risks for business go down when each risk has an owner. In an AI department, the AI director owns the solution map and the result and reports to the chief executive. The lawyer and the security team own data, 152-FZ compliance and vendor contracts.

The IT department connects AI to the CRM, the accounting system and the phone line and grants access. Department heads own their processes and their numbers: what the agent does by itself, what it passes to a person and which metric shows that everything works.

In a small business, this is one responsible person and an AI director for a few hours a week; in a large one, a separate unit. Our page on the structure of an AI department describes who is on the team and what each person owns.

Risks go into an AI solution map as a separate section: what you can give to AI and what you cannot, where to store data under 152-FZ and how not to depend on one vendor. This way AI and information security are discussed before the start, not after the first leak.

You can start with a 30-minute review. We will see which services employees already use, where the data goes and where you need a backup.

Sources and fact-checking

Checked by the editors on 13 October 2026.

  1. ComNews: only 26% of Russian companies have an AI development strategy, 22 December 2025 (comnews.ru)

Found a mistake? Tell us

Publications

Contact us