1. Architecture Platform is not an AI system
Architecture Platform is a governance platform: it does not itself run any AI system within the meaning of Article 3 of Regulation (EU) 2024/1689. It allows its customers to maintain the register of their own AI agents and applications, to classify their risk level, to record human reviews and to document decisions.
On that basis, Architecture Platform has, as a vendor, no direct obligation under Chapters II and III of the Regulation; it acts as a compliance tool for AI providers and deployers.
2. What the platform provides
2.1 Register of AI agents and applications (Articles 11, 12 and 26)
For each deployed AI system, the platform makes it possible to record:
- Identifier, business owner and tech lead.
- Vendor and underlying model (OpenAI, Anthropic, Mistral, in-house model, etc.).
- Type of agent (chat, RAG, automation, classification, etc.).
- Risk-level classification (minimal, limited, high, unacceptable) according to Annexes I and III of the Regulation.
- Sensitivity of the data processed (public, internal, confidential, PII, regulated).
- human-in-the-loop and kill-switch indicators.
- Link to the information-system applications that consume it or feed it.
2.2 Architecture reviews (Article 9 — risk management)
For each proposed AI system, the Architecture Review Board (ARB) workflow records the reviewers' opinions, the activation conditions and the re-review dates. This documents the continuous risk management expected by Article 9 of the Regulation, in particular for systems classified as “high-risk”.
2.3 Expiry and re-review policy (Article 9(2)(c))
ARB notices for AI systems come with a validity period, configurable according to the risk level (by default 12 months for HIGH_RISK, 24 months for LIMITED_RISK). On expiry, a re-review is scheduled automatically.
2.4 Audit trail of agent actions
Every creation, modification or status change of an AI agent is logged (who, what, when, old and new state) and retained for the term of the subscription. These logs can be exported in order to respond to a regulatory audit request.
3. What the platform does not do
Architecture Platform is not a substitute for the classification, DPIA and compliance work that each organisation must carry out for its own AI systems. It does not:
- Audit the models themselves (bias, robustness, explainability).
- Carry out the technical verification of training data quality.
- File declarations on behalf of its customers with the European AI Office.
4. Timeline for the application of the Regulation
Regulation (EU) 2024/1689 entered into force on 1 August 2024. The obligations apply in stages: the first three have passed and already apply, and only the August 2027 stage is still to come:
- February 2025 — already applicable: prohibitions on unacceptable practices (Article 5).
- August 2025 — already applicable: obligations on GPAI models (Articles 51 et seq.).
- August 2026 — already applicable: general applicability, including high-risk systems (Annex III).
- August 2027 — still to come: final applicability (systems embedded in regulated products, Annex I).
5. Hosting and sovereign AI
The data entered in the register — agent records, risk matrices, audit logs — remains hosted in the European Union. Architecture Platform does not transmit any of this data to third-party AI providers. Any AI assistant integrations (for example DAS suggestions) are optional, can be disabled per tenant and are explicitly flagged.
6. Regulator requests
In the event of a request for information from a competent authority (CNIL, or the national AI Act authority), Architecture Platform cooperates fully and makes the necessary exports available to its customers (register, audit trail, ARB sessions).
7. Contact
For any question regarding this positioning:
legal@arch-platform.com