Are standard contractual clauses enough for US AI services?

Standard contractual clauses are the most important instrument for sending personal data to a third country, yet a signature does not make a transfer lawful. Companies that use US AI services such as ChatGPT or Copilot need to check and document more than the contract.

Klaus Bauer5 min read

Illustration: A stamp on paper leaves a scale at zero; only a key and cogs placed beside it make the scale tip.

Why the signature alone is not enough

Under Art. 44 GDPR, personal data may only go to a third country if the conditions of Chapter V are met. One route is an adequacy decision under Art. 45, where the European Commission has found that a country offers adequate protection. Without one, appropriate safeguards under Art. 46 can apply, and the standard contractual clauses (SCCs) adopted by the Commission are the best known of these.

A recent article from Dr. Datenschutz points out that the clauses standardise contractual duties but replace neither a legal basis for the processing nor the other GDPR obligations. It also reminds readers to check first whether a Chapter V transfer exists at all. The clauses give a solid frame, and the real work lies in applying them to the actual data flow.

Which module fits which role?

The SCCs in Implementing Decision (EU) 2021/914 contain general clauses and four modules. Module 1 covers controller to controller, module 2 controller to processor, module 3 processor to processor and module 4 processor to controller. The decisive factor is the real role under Art. 4 No. 7 and 8 GDPR, whatever label the contract carries.

For a US AI provider that processes your prompts on your instructions, module 2 is commonly the starting point. One party can hold different roles in different processes, so several modules may apply between the same companies. The clauses are to be used unchanged, and extra terms are only allowed if they do not contradict them or weaken the rights of data subjects.

Filling in the annexes and describing data flows

Annex I names the parties and their roles, describes the transfer and identifies the competent supervisory authority. The description should cover groups of data subjects, categories of data, purposes, frequency, retention and any sensitive data. Annex II lists the technical and organisational measures actually in place, and Annex III lists sub-processors where relevant.

Entries such as security measures or employee data are too vague to allow a meaningful review. For AI services the annexes should reflect what staff really enter, for example contracts, HR files or customer emails. Modules 2 and 3 build in the processor duties of Art. 28(3) and (4) GDPR, so a separate processing agreement is generally unnecessary for that processing if the clauses are completed properly.

Transfer impact assessment and supplementary measures

In the Schrems II ruling the Court of Justice left the SCCs valid but made clear that they do not guarantee protection in every country. A private contract does not bind foreign authorities, so exporter and importer must assess whether the law and practice of the destination undermine the clauses. Clause 14 addresses this assessment and clause 15 covers access requests by public authorities. The documented result is called a transfer impact assessment (TIA).

The EDPB Recommendations 01/2020 describe six steps: map the transfers, identify the transfer tool, assess third-country law and practice, identify supplementary measures, take the formal steps, and review regularly. Technical measures can include encryption with keys held in the EEA or effective pseudonymisation. Organisational measures include restricted access rights, logging and escalation procedures. Contractual promises on transparency and challenging orders can add to these but do not suffice where mandatory foreign law requires plaintext access. If the clauses cannot be met even with such measures, the transfer must be suspended or ended.

For US providers, also check whether the recipient is actively certified under the EU-US Data Privacy Framework and whether the transfer falls within the certification. The source notes that this framework rests on legally uncertain ground, which is a reason to keep the SCC analysis current.

What companies should check and document now

  1. Record every data flow to AI and cloud services, including remote access and onward transfers.
  2. Determine the role of each party per transfer and assign the module.
  3. Complete the annexes with the real data categories, purposes and measures.
  4. Document the TIA, the supplementary measures and who approved them.
  5. Re-assess when purpose, data scope, recipients, sub-processors, third-country law or security architecture change.

Typical errors are the wrong module, incomplete annexes, unclear responsibilities and the assumption that a general processing agreement already covers Chapter V. Your data protection officer or counsel can judge the individual case. A Betriebsrat (works council) may need to be involved where staff data is affected.

Using AI with company documents under these rules

souveraen.ai is an AI platform for company knowledge from clouditiv GmbH in Auerbach, Germany. It runs in two modes with the same software: On Demand, hosted in Germany, and on-premise. More on this is on the page about GDPR-compliant AI.

With Private Spark the platform runs on an NVIDIA DGX Spark in your own network, and Air-Gap Enterprise works without any outside connection. Answers cite their source, and connected systems are read with the existing permissions. Holders of professional secrets can sign a confidentiality undertaking under section 203 of the German Criminal Code. If you want to keep data in-house, the page on local AI describes that route, and the security page covers the protective measures.

Sources

More articles

All articles

How safe are AI API routes via resellers and routers?

Security firm Dragos sent about two million requests with fake credentials to AI providers and found compromised services and resellers reading customer prompts. Companies that use AI through third parties need to know who sits on the path of their data.

Data protection4 min

Run Text Generation Inference with Docker on Your Server

Text Generation Inference (TGI) serves open language models from a single Docker container on your own hardware. This tutorial starts it, sends requests over HTTP and keeps prompts inside your network.

Tutorial4 min

Cloud outage liability: who answers when Microsoft fails

On 12 November 2026 a software fault took Exchange Online and Outlook offline for about 36 hours. A legal analysis published shortly afterwards argues that responsibility for such outages often lies with the operator rather than the provider.

Regulation4 min

Test it with your own documents.

Free for 3 days, then from €24.90 per user per month. Cancellable monthly.

Free for 3 days, then from €24.90 per user per month.Free for 3 days
Try free for 3 days