AI at work: which data to keep out of AI tools, the guidance from the FOCS and the FDPIC

Writing an email, translating a text, summarising a spreadsheet: for many people in Switzerland AI tools are now part of everyday work, in Ticino SMEs too. On 16 October 2026 the Federal Office for Cyber Security (FOCS), together with the Federal Data Protection and Information Commissioner (FDPIC), published a questions-and-answers page on AI in everyday working life, with practical recommendations.

The main message is not about technology but about habits: what gets pasted into a prompt and what happens to that data.

What oversharing means

The FOCS uses the term «oversharing» for the careless disclosure of information, often in good faith and to save time, without realising the data protection consequences. Simply copying the full text of an email into a prompt is enough for the personal data it contains to end up in the AI tool.

At work this is not only about one's own data. Client, patient or employee data and business secrets that should never leave the company are often involved. Health data, personnel files or lines of code containing access details for a database with personal data usually count as particularly sensitive personal data under the Federal Act on Data Protection (FADP), which carries stricter requirements.

The less visible risk: training

According to the FOCS, the biggest and least noticed risk is that many AI tools can use the data entered to train future models, unless this is excluded by contract. Once entered, information cannot be taken back and, in the worst case, it can surface in answers given to other users.

There is also a risk in the opposite direction. Texts, photos, audio and video published on a company's website and social media can be collected and exploited by fraudsters, for example for highly personalised emails, imitated voices and deepfakes used in CEO fraud.

The law already applies

As early as 2023 the FDPIC pointed out that current data protection legislation is directly applicable to AI: it is technology-neutral and covers both the training of models and the entry of a prompt. For a company the concrete risk is breaching the principles of Article 6 FADP, namely purpose limitation, proportionality and transparency. The consequences can be a notification to the FDPIC, reputational damage and, for intentional breaches of certain obligations, criminal penalties.

The FDPIC also highlights the right, under the FADP, to have an automated individual decision reviewed by a person, for example when job applications are pre-screened. On the legislative side, the Federal Council has instructed the Federal Department of Justice and Police to prepare a draft to regulate AI and implement the Council of Europe Convention on Artificial Intelligence; the consultation is expected in January 2027, with a sector-specific, risk-based approach. More background is on the FDPIC page AI and data protection.

The FOCS's four recommendations

• Assess AI-generated content critically.
• Before every input, check whether it contains personal or sensitive data, such as personal details, client data or trade secrets, and if so anonymise it.
• Where possible, use the Business or Enterprise version of the tools provided by the company rather than private accounts.
• Know when data is used for training and when it is not, and use opt-out options where they exist.

What an SME should put in place

The FOCS says companies should set clear internal guidelines and train their staff. In practice, for an SME:

• A list of approved tools, with the company accounts to use, and no private accounts for work data.
• A list of data that must never be entered: health data, personnel files, identifiable client data, confidential contracts, passwords, API keys and access details.
• Checked contracts: prefer enterprise solutions where exclusion of data from training is guaranteed by contract, and check where the data is processed.
• Technical measures where needed, such as the data loss prevention tools mentioned by the FOCS.
• Human review of results before use, especially in client-facing texts and in decisions about people.
• Care with what is published on the website and social media: voices, faces and organisation charts make targeted fraud easier.

These rules fit into the wider picture of organisational security, the weak point identified by the «SME Cybersecurity 2026» study, and go together with secure logins for the company accounts of AI tools. On the skills side, demand for AI profiles in Swiss SMEs is rising fast, as the almost 5,000 AI job ads in Switzerland show; for technical trends see also AI in the next two years.

In Ticino: a session on AI governance

For those who want to go further locally, Fondazione AGIRE and SUPSI-DTI continuing education offer a series of three evenings on the conscious use of AI in companies (in Italian), aimed at Ticino SMEs. The last session, on 17 November 2026, covers data, responsibilities, rules and governance of AI. According to the organisers, participation is free with prior registration, with a maximum of 30 participants per evening.

How ElaraTeq Sagl can help

ElaraTeq Sagl helps SMEs choose tools, write simple internal rules on AI use and integrate AI into websites, CRM and ERP systems without exposing data that must stay in the company, with hosting in Switzerland where needed. More on our consulting or via the contact page.

Sources

• FOCS, «AI in our everyday working lives: Making the most of the opportunities, protecting our data», 16 October 2026: bacs.admin.ch
• FDPIC, «Current data protection legislation is directly applicable to AI», 9 November 2023: edoeb.admin.ch
• FDPIC, «AI and data protection»: edoeb.admin.ch
• Fondazione AGIRE, «Tre serate sull'utilizzo consapevole dell'AI in azienda, in collaborazione con SUPSI DTI», 18 September 2026 (Italian): agire.ch