The EU AI Act is the world’s first comprehensive AI regulation, and it does not only apply to tech corporations. Mid-market companies that use AI in customer service, hiring, lending or production also fall within its scope. The most common question in the mid-market is therefore: what do we really have to do, and by when?
In short: The EU AI Act applies regardless of company size. What matters is the risk class of the AI system, independent of revenue. Some obligations already apply, and the high-risk obligations take effect after the Digital Omnibus from December 2027 and August 2028.
The deadlines at a glance
The Digital Omnibus moved the dates, but the roadmap is clear:
- Since February 2025: prohibited practices (Art. 5) and the AI literacy duty (Art. 4).
- Since August 2025: obligations for general-purpose AI models (GPAI).
- From December 2027: high-risk obligations for standalone systems.
- From August 2028: high-risk obligations for embedded product systems.
The risk classes
The EU AI Act ranks AI systems by risk. For readiness, the correct classification is the first and most important step:
- Prohibited practices. Unacceptable risk, such as social scoring or manipulative systems. These are banned.
- High-risk systems. Applications in sensitive areas such as HR, creditworthiness, critical infrastructure or education. These carry the most extensive obligations.
- Systems with transparency duties. Chatbots or generated content that must be marked as such.
- Minimal risk. The majority of operational AI applications, for which the AI literacy duty is the core requirement.
A readiness checklist
A pragmatic starting point for the mid-market:
- Inventory. Capture all AI systems in use and planned, including AI inside purchased software.
- Classification. Assign each system to a risk class and derive the resulting obligations.
- AI literacy (Art. 4). Role-based training for everyone working with AI, documented and repeatable.
- Governance for high-risk. Build risk management, data governance, logging, human oversight and technical documentation.
- Accountability. Clarify who owns AI governance in the company and whom it reports to.
Where the work really sits
A postponed deadline does not reduce the workload. Risk management, data governance and documentation for a high-risk system take months, tightly interwoven with how the system is built and operated. Companies that use the extra time build governance as part of delivery, well before the deadline. That is where it is decided whether AI in the mid-market makes it from pilot into reliable, auditable production.