How are AI agents addressed within the AI Act?
The European Commission’s AI Act Service Desk has provided one of the clearest statements so far on how AI agents fit within the AI Act framework.
The key takeaway is that AI agents are not a separate legal category under the AI Act. Instead, they will generally fall within existing concepts, namely AI systems and, in many cases, general-purpose AI (GPAI) models.
A typical AI agent contains a GPAI model and qualifies as an AI system because it includes interfaces and other system components. As a result, the AI Act’s existing obligations apply without the need for a dedicated regulatory regime for agents.
Several practical points stand out:
- AI agents remain subject to the prohibited practices provisions in Article 5, including restrictions on harmful manipulation and exploitation of vulnerabilities.
- From 2 August 2026, transparency obligations under Article 50 become particularly relevant where agents interact with individuals or generate content.
- If an agent qualifies as a high-risk AI system, the additional requirements of Chapter III will apply from the relevant implementation dates.
For providers of GPAI models, increasingly agentic capabilities, autonomous behaviour and tool use may be relevant factors when assessing systemic risk status and associated risk management obligations.