Five terms every European agentic commerce stakeholder must understand

European commerce platforms are deploying autonomous software agents that initiate purchases, authorise payments, and negotiate contract terms without per-transaction human sign-off. The regulatory frameworks governing these acts, principally the Commercial Agents Directive 86/653/EEC, PSD2 and the forthcoming PSD3/PSR package, GDPR, and the EU AI Act, were drafted with human actors in mind. Each framework assigns obligations and liability to identifiable legal persons, and each depends on a clear account of who instructed whom, under what authority, and when.

The five terms addressed in this brief sit at the intersection of those frameworks. When a software agent initiates a SEPA Instant payment on behalf of a consumer, the structural determinations that follow concern whether the agent was authorised, by which principal, under what scope, whether that scope was recorded in a form a court or regulator can inspect, and whether the transaction was final before any party could intervene. The answers hinge entirely on how precisely these five terms are defined and operationalised within a given system's architecture.

Under EU commercial law, grounded in the Commercial Agents Directive 86/653/EEC and its national implementing legislation across member states, an agent is a person authorised to negotiate or conclude contracts in the name of the principal. The principal is bound by acts the agent performs within the scope of that authority. Acts outside that scope create personal liability for the agent, and potentially for the principal if apparent authority can be established. This structure presupposes a human actor who can be held to account, a mandate whose terms can be interrogated, and a temporal sequence in which authority precedes action.

Delegation is the instrument that defines scope. In classical commercial law, delegation may be express (a written mandate specifying permitted acts) or implied (derived from the nature of the relationship). In agentic commerce systems, delegation must be rendered machine-readable: a permission scope specifying, for example, the maximum transaction value, the categories of goods that may be purchased, the counterparties with whom the agent may deal, and the time window during which authority holds. Each parameter of that scope is a legal boundary; acts beyond any single parameter are, in the classical framework, unauthorised.

Settlement introduces a temporal complication that the classical framework did not anticipate. When a payment agent initiates a SEPA Instant Credit Transfer and settlement occurs within the scheme's ten-second ceiling, the financial obligation discharges before any human review is possible. Classical agency law resolves disputes after the fact, examining whether the agent acted within scope at the moment of instruction. Settlement at that speed means that the financial consequence is irreversible before the dispute can even be identified. The liability question therefore turns entirely on what was recorded at the moment of delegation, because the window between instruction and irreversible consequence is too narrow for real-time intervention. Attestation is the mechanism that bridges this gap: a cryptographically verifiable record, generated when authority is granted, that allows a court, a payment service provider, or a regulator to reconstruct the scope of delegation after settlement has already occurred.

Precise definitional grounding on these five terms changes system architecture in at least three concrete directions.

First, settlement design can no longer treat payment finality as a purely technical event. Because SEPA Instant settlement is irreversible within the scheme's ten-second ceiling, the delegation record must be generated and cryptographically bound to the payment instruction before the instruction is transmitted. Platforms that treat attestation as a post-hoc audit function rather than a pre-settlement prerequisite will face disputes they cannot resolve: no record of scope means no basis for determining whether the agent acted within authority, which defaults, under most member-state interpretations of the Directive, to principal liability for the full transaction value.

Second, audit trail architecture must distinguish between the moment of delegation and the moment of settlement. These two events frequently occur hours or days apart in agentic systems, because authority may be granted once (the consumer authorises an AI shopping agent for a monthly budget cycle) and exercised repeatedly and autonomously across dozens of individual transactions. Each exercise of authority is a distinct legal act. Compliance teams must therefore structure logs such that each transaction can be traced back to a specific delegation event, with the scope at that moment recoverable in a form admissible as evidence.

Third, liability allocation in commercial contracts between platform operators, AI system developers, merchants, and consumers must explicitly assign responsibility for each definitional failure mode: unauthorised agent acts, scope ambiguity, settlement before dispute identification, and attestation record loss or tampering. The EU AI Act's provisions on high-risk AI systems introduce conformity assessment obligations that begin to address developer liability, but they do not resolve the downstream allocation between platform operator and the deploying principal. That allocation remains a matter of contract, and it must be drafted with these five structural distinctions in view.