The AI expert named the reasons for the leakage of correspondence between users of neural networks

The correspondence of users with artificial intelligence may be publicly available not only due to hacking, but also due to access control errors. Nikita Prokhorov, co-owner of the Sidorin Lab reputation management agency and an expert on artificial intelligence, told Izvestia on July 22.

According to the expert, many services allow you to create public links to dialogues for collaboration. If such access is open, search engines can index the page, making the correspondence available to outsiders.

"This is more of an access control error than a compromise of the infrastructure of the service itself. However, the consequences for the user will be the same — confidential information becomes available to outsiders," Prokhorov said.

The expert emphasized that users often trust chatbots with any information, including passport data, phone numbers, addresses, medical information, documents, financial indicators and passwords. Such data should not be shared with open AI services, especially corporate ones.

Selective model: foreign neural networks will not be banned in Russia

How will this affect the implementation of artificial intelligence in its regular users?

Using foreign platforms requires a risk assessment: companies need to know where data is stored, how its security is ensured, and whether this is consistent with domestic policy. Organizations have rules prohibiting uploading documents with trade secrets or personal data without permission.

If the correspondence has already been publicly available, it is necessary to close the link as soon as possible, contact the support service to delete the material and evaluate what information has been disclosed. If passwords or tokens are leaked, they should be replaced immediately.

Prokhorov also noted a psychological aspect: users began to perceive neural networks as a trusted colleague, rather than as an external service. Because of this, the amount of confidential information that is voluntarily shared with chatbots is growing faster than the culture of safe work with them.

Earlier in the day, it was reported that the dialogues of the clients of the Chinese chatbot DeepSeek, including working documents and personal messages, were publicly available. When using a special search query in the Google system, the correspondence is displayed in the results. Clicking on them allows you to view the names of the documents that the user sent, all the questions they asked, and the answers they received from DeepSeek.