In this article, learn when the AI Act applies to Swiss companies and what you need to keep in mind.

The AI Act generally classifies so-called AI systems based on a risk assessment and derives different requirements and regulations for the various stakeholders from this classification. Since the AI Act—like, for example, the EU’s General Data Protection Regulation (GDPR)—has so-called extraterritorial effect, Swiss companies, regardless of their size (i.e., including SMEs and startups), must comply with the AI Act in various contexts.

The AI Act is a complex set of EU regulations. In this article, we will focus on a few key aspects.

Which Swiss companies must comply with the EU's AI Act?

A company based in Switzerland must comply with the AI Act in the following situations (not exhaustive):

  • When a provider places an AI system on the market or puts it into service in the EU;
  • If a provider places a general-purpose AI model on the market in the EU;
  • If the output generated by a provider’s or operator’s AI system is used in the EU;
  • Authorized representatives of providers.

What is meant by an AI system?

An AI system is defined as a machine-based system designed to operate with varying degrees of autonomy, which, once it begins operation, can adapt and, based on the inputs it receives, derive explicit or implicit goals to generate outputs—such as predictions, content, recommendations, or decisions—that can influence physical or virtual environments.

What is meant by a general-purpose AI model?

A “general-purpose AI model” is defined as an AI model—including cases in which such an AI model is trained using a large amount of data under comprehensive self-monitoring— that has significant general applicability and is capable of competently performing a wide range of different tasks, regardless of how it is placed on the market, and that can be integrated into a variety of downstream systems or applications, with the exception of AI models used for research and development activities or the design of prototypes prior to their placement on the market.

Examples:

  • ChatGPT
  • Claude

Who is the provider?

A “provider” is defined as a natural or legal person, public authority, agency, or other entity that develops or commissions the development of a general-purpose AI system or AI model and places it on the market under its own name or trademark, or puts the AI system into service under its own name or trademark, whether for a fee or free of charge.

Examples:

  • OpenAI (ChatGPT)
  • Microsoft (Copilot)

Who is the operator?

An “operator” is defined as a natural or legal person, public authority, agency, or other body that uses an AI system under its own responsibility, unless the AI system is used in the course of a personal and non-professional activity.

Examples:

  • Any company that uses an AI system.

When is a risk considered unacceptable in an AI system, and what constitutes prohibited practices?

The following scenarios are classified as posing an unacceptable risk:

  • Subliminal influence through deliberately manipulative or deceptive techniques
  • Exploitation of weaknesses or vulnerability
  • Social Scoring
  • Predicting Criminal Activity
  • Facial Recognition Database
  • Emotion Recognition in the Workplace or in Educational Institutions
  • Biometric Categorization
  • Biometric Monitoring

Examples:

  • Targeted manipulation of social media content to advance political goals.
  • Analyzing social media data to assess whether someone is creditworthy.

What are the consequences if an AI system poses an unacceptable risk?

AI practices are completely prohibited if they are classified as posing an unacceptable risk. They are therefore also referred to as prohibited practices.

When is an AI system considered high-risk?

On the one hand, if the product or a safety component of a product is subject to EU harmonization regulations (e.g., medical devices, industrial machinery, toys, aircraft, automobiles). On the other hand, use cases as specified in Annex III (subject to change or modification):

  • Biometrics:
    • biometric remote identification systems
    • Biometric categorization based on sensitive attributes
    • Emotion Recognition Systems
  • Critical Infrastructure:
    • Security systems for the management and operation of infrastructure such as transportation, water, and energy supply
  • Education:
    • Systems for Access, Assessment, and Monitoring in Educational Institutions
  • Employment and Human Resources Management:
    • Systems for hiring, performance evaluation, working conditions, and monitoring of employees
  • Basic Services:
    • Assessment of Eligibility for Public and Private Benefits (e.g., health care, loans)
  • Law enforcement:
    • Risk Assessment, Lie Detectors, Evidence Evaluation, Profiling
  • Migration and Border Control:
    • Lie detectors, risk assessments, assistance with asylum and visa applications
  • Law and Democracy:
    • Support for judicial authorities, influencing election results

What are the consequences if an AI system is classified as high-risk?

High-risk AI systems are subject to strict and comprehensive regulations.

Providers of such high-risk AI systems must, for example, comply with the following obligations:

  • Quality and Risk Management System
  • Data Governance
  • Technical Documentation
  • Record-keeping Requirements
  • Transparency Requirements
  • Human supervision
  • Accuracy, Robustness, and Cybersecurity
  • Corrective Actions
  • Declaration of Conformity

Operators of high-risk AI systems must also comply with extensive obligations, such as:

  • Ensuring that the AI system is used in accordance with the operating instructions (technical and organizational measures)
  • Human supervision
  • Monitoring and Disclosure Requirements
  • Retention of Records
  • Fundamental Rights Impact Assessment

What transparency requirements apply to all AI systems?

Regardless of a risk classification, the following transparency requirements must always be met:

  • Providers of AI systems must ensure that individuals are informed when they interact directly with an AI system, unless this is already obvious.
  • Providers of AI systems that generate synthetic audio, image, video, or text content must ensure that the AI system’s output is labeled in a machine-readable format and is identifiable as artificially generated or manipulated.
  • Operators of emotion recognition systems or biometric categorization systems must inform individuals about the operation and processing of personal data.
  • Operators must label deepfakes—that is, AI-generated or manipulated image, audio, or video content that resembles real people, objects, places, institutions, or events, or that would falsely appear to a person to be genuine or truthful.
  • Operators of an AI system that generates or manipulates text published to inform the public about matters of public interest must disclose that the text was artificially generated or manipulated.

What applies to all providers and operators of AI systems?

Both providers and operators of AI systems must ensure what is known as “AI competence,” meaning they must ensure that their staff and others entrusted with the AI systems have sufficient knowledge of how to use AI systems.

Here is an overview of training and continuing education programs.

What are the risks of not complying with the AI Act?

Depending on the violation, the following levels of sanctions are provided for:

  • 35 million euros, or 7% of global annual revenue
  • 15 million euros, or 3% of global annual revenue
  • 7.5 million euros, or 1% of global annual revenue

When do the obligations under the AI Act take effect?

The AI Act has been in effect since early August 2024. However, the implementation requirements will be phased in as follows:

Effective February 2, 2025:

  • Ban on AI Systems That Pose Unacceptable Risks
  • Proof of AI proficiency is necessary for everyone

Effective August 2, 2025:

  • Regulations for General-Purpose AI Models

Effective August 2, 2026:

  • All regulations, with the exception of a few high-risk AI systems

Effective August 2, 2027:

  • Regulations for Certain High-Risk AI Systems

When does the AI Act not apply?

The AI Act does not apply under the following circumstances:

  • Military and security-related purposes
  • International Cooperation in the Field of Law Enforcement and Judicial Cooperation
  • Research and Development
  • Personal Use
  • Open-source AI (provided it does not involve prohibited AI practices, a high-risk system, or deepfake generation)

Summary & Assessment Rubric

The AI Act is a complex set of regulations, the specific details of which will only become clear over time—that is, as relevant guidelines are published and practical experience is gained.

In this post, we've focused on a few key points.

In practice, we recommend following these testing steps:

  1. Are we dealing with an AI system or a general-purpose AI model?
  2. If so: In which capacity, as defined by the AI Act, are you operating? Specifically, are you a provider or an operator of an AI system?
  3. Does the AI Act apply if you're outside the EU?
  4. Which risk category does the AI system fall into?

Last but not least, it should be noted that operators must also comply with the obligations set forth by the manufacturer of an AI system or a general-purpose AI model in its terms of use. These obligations may go beyond those defined in the AI Act.

Overview of training and continuing education programs.

Source Information: This article was originally written by Caroline Danner, attorney and founding partner at ONLAW, and published on her blog. We are publishing it here with her permission.