Advisory IT & Digitalisation

Can AI really be implemented and used safely in a regulated environment?

By: Christopher Oehri

The answer is ‘yes’ – provided you know which AI applications are actually already in use within your organisation. Regulated companies, particularly small and medium-sized ones, are already making extensive use of AI – in some cases without explicit approval from the executive board or the board of directors, or even without their knowledge. This is precisely where the risk for your organisation lies.
Contents

The reality: AI has long been in use – just not always authorised

In many companies, AI is already part of everyday work – without the responsible management body being aware of it. One key reason is that many employees, especially younger ones, are not prepared to forgo technological progress, even where internal guidelines do not permit it. Employees benefit immediately from the efficiency gains AI delivers, while the company bears the risk. This phenomenon is often referred to as ‘shadow AI’: the use of AI tools outside the authorised IT environment.

As a result, data – sometimes including confidential information such as client data and personal data – is repeatedly processed using AI tools that the company has neither officially introduced nor authorised.

 

Why this poses a significant risk to regulated institutions

This situation poses a considerable risk to the organisation itself, as requirements relating to data protection and confidentiality can no longer be met. The following areas are particularly affected:

  • Data protection: Entering personal data into an AI system that is not a closed system may result in this data being permanently disclosed to third parties, thereby breaching data protection law.

  • Professional and business secrecy: Entering client data into tools that are not closed systems may, without the client’s consent, breach confidentiality obligations and contractual agreements.

  • Outsourcing and third-party risks: AI tools are often operated by external, sometimes foreign, providers. The supervised institution nevertheless remains responsible for complying with outsourcing requirements.

  • Quality and traceability: AI results are not 100 per cent reliable. Without documented human review, errors may go undetected until a late stage and, in the worst case, become public.

  • Governance gap: The governing body cannot manage or take responsibility for what it is not aware of – and in a regulated environment, this is precisely the critical point.

 

The regulatory framework in Switzerland and Liechtenstein

Unlike the EU, which already has specific AI legislation in the form of the EU AI Act, Switzerland does not yet have such legislation – but it is on the horizon: a consultation draft on the regulation of AI is expected by the end of 2026. However, this is no reason to sit back and wait until then.

The situation in Liechtenstein is different: the EU AI Act already provides a comprehensive, risk-based legal framework in Europe. As Liechtenstein is a member of the EEA, the regulation has been classified as EEA-relevant; once incorporated into the EEA Agreement, it will be directly applicable in Liechtenstein.

For institutions operating in both jurisdictions, this means that while the requirements in the two countries differ in detail, the expectation remains the same: documented AI governance, endorsed by the highest governing body.

 

Our services

Our specialists at Grant Thornton Switzerland/Liechtenstein are familiar with the various requirements our clients must meet in both Switzerland and Liechtenstein. We would be happy to advise you on how to implement an AI tool in compliance with the applicable regulatory requirements, or help you determine whether tools that are not permitted are already in use within your organisation. In a structured workshop, we provide you with an overview of the current state of your AI governance, identify regulatory and operational areas for action, and highlight which measures should be prioritised for your organisation. The result: concrete recommendations and a clear implementation plan.

 

Conclusion

AI can be used safely in a regulated environment – provided its use is known, authorised, documented and monitored. Organisations that analyse and regulate their use of AI at an early stage not only ensure regulatory compliance, but also strengthen the trust of clients, business partners, auditors and supervisory authorities.