Are you pasting data into AI? Companies make the same mistake

In many organizations, artificial intelligence has ceased to be a technological curiosity and has become an everyday work tool. The problem is that the pace of its implementation clearly outpaced the creation of security and data protection procedures.

The latest analyzes show that in industries with high implementation potential, 23 percent companies use AI in business processes, and another 13 percent is just planning its implementation or sees real potential in it. Most often, technology supports work with documents and data, customer service, planning and automation of repetitive tasks.

The trend is even more visible in the legal sector. As much as 92 percent lawyers declare to use at least one AI tool. At the same time, almost half of them point to data privacy and cyber threats as key challenges related to the new technology.

– Companies are already using AI. In many organizations, the first step was not a grand strategy, but simple tasks: summarizing the contract, translating the document, preparing a response to the client. Therefore, the most important question today is not whether to use AI, but what data is sent to it – notes Wojciech Chmiel, co-creator of Anonimizer, in an interview with “Wprost”.

Risk does not start in the model, but in the document

In the debate about artificial intelligence, we most often talk about algorithm errors. Meanwhile, the real threat appears much earlier – when data is transferred to AI tools.

In companies, this is often very sensitive information: customer and employee data, contracts, offers, business strategies and correspondence. In law firms, there is also professional secrecy and responsibility for documents provided by the client.

The stakes are high. Violations of GDPR may result in fines of up to EUR 20 million or 4%. the company’s global turnover. For many organizations, however, the loss of reputation and trust of business partners can be equally severe.

– In our opinion, we need a simple standard of work: first data hygiene, then AI, because the safest data in the cloud are those that have never been sent there – says Wojciech Chmiel.

Anonymization as a filter against artificial intelligence

One approach to reducing risk is to pseudonymize data before it is sent to the AI ​​model. In practice, this means removing or masking sensitive information before it reaches the external system.

This is where Anonymizer works – a tool that operates locally on the user’s computer and prepares documents to work with AI. The original files do not leave the device, and their “cleaned” version is sent to the model. Once you’ve finished your work, you can automatically restore your data to your local environment.

The solution is addressed primarily to companies working with large numbers of documents – law offices, legal departments and business teams. Typical applications include translations, summaries, analyzes and preparation of draft documents.

– Local processing and human control are key in Anonymizer. The system detects and masks data, but the user can see the marked fragments and can approve, reject or correct them. The technology is intended to shorten the time spent working on documents and not to take away the responsibility of the person who uses them, says Jakub Dolata, co-creator of Anonimizer.

AI in companies: between innovation and the so-called shadow AI

Experts from Anonimizer point out that completely blocking AI tools in organizations often does not work in practice. Employees still reach for unofficial solutions, creating the so-called shadow AI, i.e. the use of models outside the company’s control.

Therefore, it is crucial not to limit access to technology, but to organize the rules for its use – from data classification, through security procedures, to results control.

In this approach, AI is no longer just a productivity tool, but an element of a broader information management system in which the first line of defense is the way data is prepared.