A revised deadline for one set of AI obligations does not answer which duties apply to a particular workflow today. Enterprise leaders need a system-by-system view of purpose, geography, legal role, and operating evidence. The dates below reflect European Commission guidance checked on September 5, 2026; the explorer is an orientation aid, not a legal classification of your organization.
Separate the obligation tracks
The Commission’s AI Act overview distinguishes several application dates. Prohibited-practice and AI-literacy provisions began applying on February 2, 2025. General-purpose model obligations followed on August 2, 2025, with transitional provisions. Article 50 transparency obligations apply from August 2, 2026, subject to specific exceptions and transition conditions.
The Commission’s implementation FAQ describes revised high-risk timelines of December 2, 2027 for Annex III uses and August 2, 2028 for relevant product-embedded systems. Those later dates are not a blanket pause on the Act. Keep each obligation and its applicable transition in a separate row of the register so one headline cannot silently replace several different decisions.
Identify the role before assigning the task
Providing an AI system and deploying one are different legal roles. A company can have more than one role across its activities, and a provider relationship with a vendor does not by itself settle the customer’s obligations. Map what the organization develops, brands, puts into service, and uses, including how outputs reach people in the EU.
For an initial review, record the intended purpose, affected users, relevant markets, vendor and contract, launch date, and the person responsible for the workflow. Use that record with qualified counsel to determine scope and classification. The practical value of an inventory is that it gives the legal review a defined system to assess. A generic list of AI tools cannot explain how any one tool is being used.
Put transparency into the actual experience
The Commission’s Article 50 FAQ distinguishes user notification, machine-readable marking, and deployer disclosures for specified content and uses. It also describes a limited December 2, 2026 transition for Article 50(2) marking obligations on systems placed on the market before August 2, 2026. That is not a general grace period for every transparency duty.
Translate the legal determination into a user experience and an owner. Decide where a notice appears, who supplies marking, how content is reviewed, and which team checks that the requirement survives a redesign or integration change. Keep a representative record of the released experience. A statement in a policy file is weak evidence if the customer-facing workflow does something different.
Build a register that produces decisions
For each in-scope workflow, connect the relevant duty to an accountable person, a concrete control, evidence, and a review trigger. Useful triggers include a change of purpose, an expanded user group, new data, a new model or feature, and a revised legal requirement. Distinguish an unresolved classification question from a confirmed control gap; they need different next actions.
In Meridian’s illustrative review, a customer assistant and an employment-related screening use would not be placed on one undifferentiated “AI compliance” timeline. Each would receive its own role and purpose analysis. The executive decision is which evidence and remediation to fund now, and which specialist questions must be resolved before further deployment. The record should make those choices visible without claiming that completion of a checklist establishes legal compliance.