Public AI chats have become an easy way to get everyday work done faster. Employees use tools like ChatGPT to summarize documents, improve emails, translate text, analyze information, work with code or turn meeting notes into something useful.

The productivity benefit is obvious. The less obvious question is what happens to company data along the way.

According to 2026 data from a report by Accenture, in fact, Switzerland ranks first in Europe in the integration of AI into daily workflows: 49% of Swiss workers use AI tools on a daily basis (compared to a European average of 28%), 88% of employees have full or moderate access to AI tools provided directly by their company, and 72% of workers view the impact of AI on their jobs in a largely positive light.

Despite this widespread adoption, however, citizens and workers are showing declining trust and growing privacy concerns: the sense of security felt by the Swiss while browsing the web has dropped from 5.7 to 5.3 points over the past three years. And perceptions of data security vary greatly among the cantons: in Ticino, 59.6% of respondents consider protection to be “good,” while that figure drops to 54% in German-speaking Switzerland and to just 43.5% in French-speaking Switzerland.

According to reports from the Federal Data Protection Commissioner, reports of privacy violations have more than doubled due to the challenges posed by data processing in AI.

What are public AI chats?

Public AI chats are generative AI tools that users can access directly through a public interface, typically in a browser or app. Unlike an AI solution specifically implemented within a company's controlled environment, employees can start using these tools independently and within seconds.

That accessibility is part of what makes them useful. But it also creates a new data protection challenge for companies.

What company data can enter an AI chat?

Consider a simple request:

“Summarize this document and give me the five most important points.”

The prompt itself contains nothing sensitive, but the document might. It could contain customer names, email addresses, personal information, financial figures, contract details, internal project information, or other confidential company data. The same applies to an email being rewritten, meeting notes being summarized or source code being analyzed.

In many cases, sensitive information isn't what the employee wants the AI to process. It's simply part of the context needed to complete the task.

Why isn't a company AI policy enough?

Clear policies and regulations for the use of public AI chats, or any AI tools, are important. This is crucial to combat the so-called “Shadow AI” phenomenon, that is the unregulated and often uninformed use of free tools through personal accounts, which could pose risks to corporate data.

A company can define which tools employees may use, which information is considered sensitive, and what should never be entered into an external service.

But policies still depend on people applying those rules correctly every time they use an AI chat. That becomes difficult when sensitive information is buried inside long documents, emails, datasets or other unstructured content. Telling employees not to share sensitive data is therefore an important organizational safeguard, but it doesn't have to be the only one.

Should companies block public AI chats?

Blocking access can reduce one type of risk, but it also removes tools employees use to work faster. There is another question worth asking: can employees use public AI chats while sensitive company data stays protected?

The ideal approach is to protect the information before it reaches the AI chat. Sensitive elements can be detected and replaced with anonymized values before the content is sent. The AI can then work with the context it needs without receiving the original sensitive information. For employees, the easy, frictionless workflow remains. Most importantly, though, for the company, the sensitive data doesn't travel with the prompt.

A privacy layer for public AI chats, how Artificialy solved the problem

We’ve built a solution acting as a privacy layer: eclipsery protects sensitive information before it is shared with public AI chats. Available as a browser extension and web app, it automatically anonymizes sensitive information and can de-anonymize the resulting content when needed. The process is designed to happen without changing the way people already work with AI chats. What’s most important for us and what we want to ensure as a provider: eclipsery does not store the sensitive information it protects.

The idea is simple: We want to enable companies to continue using the tools that help them work, while keeping sensitive data out of the conversation.