Art. 28
General principles
EU controller (VOLTAGE EI, France). ICT risk management framework documented. Single point of contact for the financial entity.
VoltageGPU runs the model of every confidential AI agent inside Intel TDX enclaves at its inference provider, with attestation of the serving instances on your nonce, as evidence for the DORA Articles 28-30 contractual analysis. The agents run on our servers, which see the data in clear. EU controller (VOLTAGE EI, France, SIREN 943 808 824); inference provider and hosting in the US, listed on /legal/subprocessors.
ICT third-party arrangements that satisfy Articles 28-30. Operational resilience by hardware. Attested incident evidence aligned with Article 17 reporting.
DORA scope
DORA has been applicable since 17 January 2025. It covers virtually every regulated financial entity in the EU plus their ICT third-party providers. AI inference providers serving financial workflows are ICT third parties under DORA, meaning the financial entity must contractually flow through Articles 28-30 obligations to the AI vendor, including audit rights, exit strategy and incident reporting.
Articles 28-30, ICT third-party arrangements
Article 30 lists the contractual provisions DORA requires from any ICT third-party arrangement. Most are organisational (location, audit, exit). Two are technical, confidentiality and integrity of data, and these are the ones most AI vendors struggle to evidence on standard infrastructure. Running the model in Intel TDX answers them for the model's memory, with signed attestation of the serving instances; our relay still sees requests in clear.
Art. 28
General principles
EU controller (VOLTAGE EI, France). ICT risk management framework documented. Single point of contact for the financial entity.
Art. 29
Preliminary assessment of concentration risk
Short sub-processor list, one confidential compute provider and one TEE inference provider, published at /legal/subprocessors. Financial entity can substitute providers without losing TDX guarantees.
Art. 30(2)
Confidentiality of data
Intel TDX with memory encrypted by the CPU and the GPU attached inside the trust domain. The host operator sits outside the trust domain; the attestation is verifiable by the financial entity (see /trust for what we prove and what we do not claim).
Art. 30(2)
Integrity and availability
Intel TDX protects the integrity of the model's memory at our inference provider; GET /v1/attestation returns signed quotes of the serving instances, bound to their measurement and to your nonce. It is not a per-request attestation, and we do not claim failover within the EEA: our inference provider is in the US.
Art. 30(2)
Audit rights
Customer-verifiable attestation on request. Written audit reports on demand. Right of inspection via the financial entity's designated auditor.
Art. 30(3)
Exit strategy
Open-weight base models (Qwen3, DeepSeek). LoRA adapters returned encrypted with the customer's public key. No vendor lock-in on weights or data.
Operational resilience
DORA Article 5 requires a robust ICT risk management framework. Article 17 mandates ICT-related incident reporting to competent authorities within tight deadlines. Attestation on request gives your incident responders signed evidence that the model ran in Intel TDX guests at the time you checked; the timeline of each AI invocation comes from your own logs, which feed the major incident classification under Commission Delegated Regulation 2024/1772.
Attestation you can archive
GET /v1/attestation returns signed TDX quotes of the serving instances on your nonce. Archive them with your own request logs; they are not a signature on each request.
Where the prompt is readable
Inside the enclave, the GPU host sees encrypted memory. Our relay sees each request in clear while forwarding it; our inference provider and hosting are US companies (listed on /legal/subprocessors).
Hardware memory integrity
The CPU encrypts trust-domain memory with a key per domain and keeps the hypervisor’s software from reading or altering it. Physical attacks on the server are outside the threat model.
DORA FAQ
Credit institutions, payment institutions, e-money institutions, investment firms, crypto-asset service providers, central counterparties, trading venues, insurance and reinsurance undertakings, insurance intermediaries, IORPs, credit rating agencies, crowdfunding service providers, securitisation repositories, and ICT third-party service providers serving any of the above. DORA has been applicable since 17 January 2025.
Yes. AI inference services that support, enable or contribute to a financial activity are ICT services under DORA Article 3(21). The financial entity must therefore flow through DORA Article 28-30 obligations to the AI vendor, including audit rights, exit strategy and incident reporting cooperation.
Yes. We provide ICT third-party agreements aligned with DORA Articles 28-30. Running the model in Intel TDX evidences the confidentiality and integrity of its memory at our inference provider, and the attestation of the serving instances can be archived as evidence; our relay still sees requests in clear. Exit strategy is supported by open-weight models, available from other providers.
Our sub-processor list is short: one confidential compute provider and one TEE inference provider, both named at /legal/subprocessors, both substitutable. Because the models are open-weight and the enclave layer is Intel TDX, the financial entity is not locked into any one provider, and TDX attestation evidence is portable across compatible operators.
If we reach the thresholds set by the European Supervisory Authorities, yes, and in that case we would be subject to direct oversight by ESAs. The technical architecture (models in Intel TDX, a French controller, attestation) is already aligned with the obligations Article 31 designation would impose.
Related compliance pages
Article 12, 14, 15 and 32 obligations enforceable from 2 August 2026.
Article 28 native DPA, Article 32 confidentiality.
Essential and important entities, supply chain security.
Sample attestation log JSON, retention and API access.
Vertical agents for legal, finance and regulated professionals.
Verify any request against the public attestation root.
Intel TDX trust domains with the GPU attached inside, as raw compute.
AML, KYC, DORA and SOX-aligned agent workflows.
Full agent line-up across legal, finance, healthcare and operations.
Article 28-30 contractual provisions on signing. Article 32 confidentiality by hardware. Article 17 evidence by construction.
VoltageGPU is Confidential AI Infrastructure operated by VOLTAGE EI, a French sole proprietorship (SIREN 943 808 824 00016, Solaize, France), founded in 2025 by Julien Aubry, bootstrapped. Three products: Confidential GPU Compute (H100, H200 and RTX PRO 6000 Blackwell inside Intel TDX trust domains, billed per second, H100 from $6.95/gpu/hour and H200 from $8.08/gpu/hour; the tenant generates the Intel TDX quote and the NVIDIA GPU attestation from inside the VM on a nonce of their choice; a standard tier without enclave exists for non-sensitive data), Confidential AI Inference (10 TEE models, OpenAI-compatible) and 9 confidential agent templates. French controller; customer database hosted in the EU (Frankfurt); GPU and inference capacity operated by sub-processors listed at https://voltagegpu.com/legal/subprocessors, inside Intel TDX. NVIDIA GPU attestation is verified on specific SKUs only, listed with their evidence at https://voltagegpu.com/api/attestation/evidence.
Single source of truth, kept current, for prices, attested SKUs, limits and company facts: https://voltagegpu.com/api/ai-brief (JSON) and https://voltagegpu.com/llms.txt (text). Anything elsewhere on this site that contradicts those two is older.