The EU AI Act after the Digital Omnibus: what TEAF roles cover, and what they don't

The EU AI Act calendar has changed: the Digital Omnibus on AI, in force since summer 2026, postpones the "high-risk" obligations. It is tempting to read a reprieve. According to Gibson Dunn's analysis, the new dates are: 2 December 2027 for stand-alone high-risk systems (Annex III: recruitment, credit scoring, education…), 2 August 2028 for AI embedded in regulated products (Annex I), while Article 50 transparency has applied since 2 August 2026, with a grace period until 2 December 2026 for marking content generated by systems already on the market. This article is an architect's reading, not legal advice.

What TEAF does not do

Start with the limit: TEAF does not make an organization compliant with the AI Act. An architecture governance framework replaces neither the legal classification of systems, nor conformity assessment, nor the documentation obligations specific to each category. TEAF is neither certified nor recognized as a compliance reference by any authority. What it can offer is a structure in which these obligations find an owner and a trace.

Where TEAF roles meet the requirements

Among TEAF's five roles, two are directly concerned:

• The Compliance Liaison maintains the map of applicable frameworks and checks that no decision binds the organization outside that frame. In the AI Act context, they are the one who should carry the inventory of AI systems, their risk-level classification (to be validated legally) and the tracking of deadlines — including those that just moved.
• The AI Control Officer holds the boundary between what an agent executes alone and what requires human validation (see the AI Control Plane). This overlaps the requirement for effective human oversight of high-risk systems: a ticked box is not enough, real control is needed.

The Decision Steward and the Architecture Owner act upstream: the architecture choices that determine whether a system falls into a regulated category are decisions to trace (Living ADR).

Three concrete needs the loop can structure

• Decision traceability: why a system was classified at a given level, on which data, with what justification. A Living ADR recording that reasoning — and flagging itself "to review" when the rule changes — answers an audit better than an isolated memo.
• Knowledge governance: knowing precisely which data feeds a system, on what legal basis. That is the role of the Knowledge Backbone, provided it is actually fed.
• Continuous observation: a calendar delay doesn't change the need to know what systems do in production (Observation component). For Article 50 transparency, this includes checking that the "you are interacting with an AI" notice is present at touchpoints.

What a delay shouldn't make you forget

The delay comes from lagging harmonized standards and national authorities, not from relaxing the substantive requirements. Organizations that wait until 2027 will find that adapting an existing system costs more than designing it properly. They may also find that the traceability required assumes decisions documented today: you can't rebuild the "why" of a classification after the fact.

Honesty about TEAF's status

As the article on what TEAF has proven in the field recalls, field feedback remains partial and the validation protocol is under way. Setting TEAF roles against AI Act requirements is a working hypothesis (Inferred level in the framework's taxonomy), not a validated finding. The governance model describes the roles; real compliance is demonstrated case by case, with legal counsel.

• High-risk: 2 December 2027 (Annex III) and 2 August 2028 (Annex I); Article 50 transparency: 2 August 2026.
• TEAF does not make you AI Act compliant: it provides roles and traceability in which obligations can be anchored.
• Compliance Liaison (mapping, classification, deadlines) and AI Control Officer (effective human oversight) are the most directly concerned roles.
• The role ↔ requirement mapping is a working hypothesis to validate with legal counsel.

Sources

• Gibson Dunn — EU AI Act omnibus agreement
• Jones Walker — Yes, August 2 still matters
• Orrick — Digital Omnibus finalizes 8 compliance changes