Starting 2 August 2026, the European Commission's AI Office and national authorities begin enforcing the AI Act. New transparency rules require AI systems to disclose their nature to users, and AI-generated or altered content must be labelled and machine-readable. Swiss organisations building or deploying AI systems—especially those with EU customers—need to understand the enforcement structure and their obligations.
On 2 August 2026, the European Commission’s AI Office and national authorities began enforcing the Artificial Intelligence Act. For Swiss organisations that develop AI systems, use general-purpose AI models, or serve EU markets, compliance is now immediate. Transparency rules and prohibited practices are live; high-risk AI obligations remain postponed until December 2027.
What is being enforced right now
Transparency obligations. Interactive AI systems must disclose they are AI, not human. Any AI-generated or altered content must be labelled, and deepfakes must carry explicit labels. Machine-readable marks must be embedded in AI-generated content for automated detection. Purpose: reduce deception and manipulation.
Prohibited practices. The AI Act bans systems designed to manipulate people, exploit vulnerabilities in specific groups, or unfairly score individuals in ways that threaten their rights. This includes systems that create discriminatory outcomes based on protected characteristics.
GPAI provider obligations. Providers of large language models and foundation models must document their training data, publish sufficiently detailed summaries of training content, and implement copyright policies. Providers of advanced GPAI models (those that may pose systemic risks) face additional requirements to address risks of large-scale harm, including cyber incidents, loss of control, and threats to critical infrastructure.
Who enforces the rules
The EU AI Office enforces rules for GPAI providers and AI systems they offer. National competent authorities enforce transparency and prohibited-practice rules for other AI systems. A Swiss organisation offering AI tools to EU users may face enforcement from multiple national authorities depending on jurisdiction and use case.
What this means for Swiss organisations
If you offer AI systems to EU customers: Transparency compliance is now required. Interactive systems must disclose their nature; AI-generated or altered content must be labelled and machine-readable marked.
If you use or resell GPAI models: You remain liable for prohibited practices (unfair scoring, manipulation). Integrate GPAI models with documented governance to avoid enforcement for transparency violations.
If you develop GPAI models: Direct obligations to the Commission now apply—documentation, training data summaries, and copyright policies are compliance requirements, not best practices.
Enforcement is live: The AI Office has established complaint and whistleblower tools. Compliance is the prudent assumption from day one, even as national authorities build capacity.
Timing ahead
The AI Omnibus postponed the application of the rules on high-risk AI systems to 2 December 2027, and postponed rules for high-risk systems integrated into regulated products to 2 August 2028. New prohibitions on non-consensual intimate imagery and child abuse material apply from 2 December 2026.
To keep on top of things, readers can refer to mavai.ch/en/timeline/ to track regulatory enforcements.
Next steps
- Audit for transparency compliance: Interactive systems must disclose AI nature; generated or altered content must be labelled and machine-readable marked.
- Review for prohibited practices: Assess systems for unfair discrimination and manipulation risks.
- Document GPAI usage: Record how you use foundation models, what data you process, and what governance controls you have.
- Build an evidence trail: Document due diligence and controls; this is what enforcement audits will examine.
Compliance is immediate. Swiss organisations serving EU markets should act now.