AI and Data Transfers to US Providers

Companies may use AI services from US providers when every data flow has a lawful transfer mechanism and a suitable contract. Language models process inputs on servers whose location and access rights an organization must understand before the first employee sends a prompt. For executives and business-unit leaders this is a governance task: which application receives which data, and under what conditions? This article provides an overview and does not constitute legal advice.
What counts as a transfer to a third country?
A transfer to a third country occurs as soon as personal data reaches a recipient outside the European Economic Area or can be accessed from there. For AI services this covers every input including uploaded documents, the generated response as well as the log data a provider stores for operations and abuse detection. Remote access by a US support team to data held in a European data center also counts. The rules are set out in Chapter V of the General Data Protection Regulation. The article Data protection and AI covers how language models can be used in line with data protection more broadly.
Today's practice stems from the case law of the Court of Justice of the European Union. The “Schrems II” judgment of 2020 invalidated the former Privacy Shield and has since required companies to assess and safeguard transfers themselves. With AI this weighs more heavily than with conventional software, because inputs are freer and more sensitive. A sales representative pastes a customer email into the input field, a controller uploads a staff list, and both happen without a ticket or an approval.
A storage location in Europe does not end the review. The US CLOUD Act of 2018 gives US authorities access, under certain conditions, to data controlled by a US company, even when it is stored in Europe. An EU region therefore reduces the risk, yet it does not replace the transfer mechanism.
Transfer mechanisms for US providers
Two mechanisms are available for transfers to the United States, and they complement each other. The first is the European Commission's 2023 adequacy decision on the EU-US Data Privacy Framework. It permits transfers to US companies certified under this framework without additional safeguards. Certification applies per company, and the US Department of Commerce publishes the list of certified organizations. Check the provider's status during procurement and document it with a date. Make sure the certification covers the legal entity with which you sign the contract.
The second mechanism is the standard contractual clauses, which the Commission revised in 2021. They bind provider and customer to protective obligations. They apply where no certification exists and serve, with certified providers, as a contractual fallback in case the adequacy decision falls. Clause 14 of the standard contractual clauses obliges the parties to carry out a documented transfer impact assessment. It evaluates whether the law of the destination country undermines the agreed safeguards and names the supplementary measures.
Both mechanisms govern only the transfer. The contract governs the provider's role as processor and the retention periods, as well as sub-processors and whether customer data is used to train models.
Is the EU-US Data Privacy Framework still valid?
The Data Privacy Framework is valid and provides a sound legal basis for transfers to certified US companies. It applies until the Court of Justice annuls the adequacy decision or the Commission amends it. The Commission reviews the decision periodically; the first review in 2024 confirmed that the US safeguards are functioning.
Legal challenges are part of the framework's nature, as with both of its predecessors. The core of the criticism is once again access by US intelligence services. The United States responded with a 2022 executive order that set proportionality limits and created a dedicated redress court for EU citizens. Whether these safeguards hold before the Court of Justice is ultimately for the Court to decide. For companies this leads to a clear rule: use the framework and secure yourself contractually at the same time.
What happens if the Data Privacy Framework is struck down?
If the Court of Justice annuls the decision, transfers that rely solely on the framework lose their legal basis with immediate effect. The “Schrems II” judgment provided no transition period. Companies whose contracts already contain standard contractual clauses with a transfer impact assessment continue on that basis and update the assessment. Companies without this fallback must renegotiate contracts under time pressure or switch applications off. Precaution costs little today: include the standard contractual clauses in every contract with a US provider, even if it is certified.
How do you assess a US AI provider?
You assess a US AI provider with five fixed questions that stay the same for every application. A uniform review framework gives business units an answer in days rather than months, because not every application starts as a special case.
- Data class: which data flows into the application?
- Data flow: where does the data go, who accesses it, where are the logs?
- Transfer mechanism: is there a certification, are standard contractual clauses in place, and is both documented?
- Technical safeguards: which region, which encryption, which filters before hand-over?
- Operations: who owns the application, and when is it reviewed again?
The answer to the first question determines the effort for the rest. Public content goes through a shortened review. An application with customer data receives the full review with data protection and information security involved. The data class also sets the sequence: applications with low data risk and high value start first, the sensitive ones follow.
Technical design of the data flow
Technical measures reduce the volume of data that leaves the European legal area in the first place. Many providers offer processing in an EU region, and this option belongs in both the configuration and the contract. An intermediate gateway replaces names or account numbers before the request reaches the provider and restores them in the response. With applications built on the company's own documents, the knowledge base stays in the company's own system and only the relevant extract goes to the model.
If these measures are not enough for a data class, the remaining option is to run a model on European or company-owned infrastructure. The decision follows the business case: it weighs the extra effort of self-hosted operation against the value of the application and of the data.
Chapter V of the GDPR does not apply to data without personal reference. The EU Data Act does, however, oblige providers of cloud services to protect such data against unlawful access by third-country authorities. Confidentiality agreements and information security rules apply in addition. A review framework that covers both kinds of data spares business units two separate procedures.
From approval to ongoing control
An approval applies to the state that was reviewed. Providers change sub-processors or move processing regions, and each of these changes can alter the data flow. Operations therefore need a register of approved applications with data class and transfer mechanism as well as a named contact. Add to this a fixed review cycle, at least annually and whenever sub-processors change, and a route by which business units register new applications.
The procedure works when it is faster than the alternative. A business unit that receives an answer within a few days registers its application. A business unit that waits for months uses a personal account, and the organization loses sight of its data flows. The article From pilot to scale shows how the review fits into the wider rollout.
Niehues Consulting as a partner for implementation
The legal assessment lies with the legal department and specialist lawyers. Turning that assessment into a working procedure is a delivery task, and this is where Niehues Consulting comes in. The approach follows the Niehues Delivery Model with the phases Understand, Build and Roll out. Three of its checkpoint questions determine the pace for data transfers. Who decides, who delivers, who uses? Does the prototype prove its value in the real process with real data? And who is accountable for operation from day one?
In AI management, business analysis captures the applications with their data flows, formulates requirements and assesses the value in the business case, so that selection and sequence follow the data classes. The prototype shows in the real process which data an application actually needs, and thereby fixes the data class that sets the review effort. Rollout defines processes and responsibilities so that register and registration route work in daily business. Delivery management steers the interplay of business unit and procurement with data protection and IT until the application is in operation.
Founder Marin Niehues is a management consultant who brings AI into operation, in projects ranging from VC-funded scale-ups to DAX 40 corporations. The entry point is a free initial consultation of 30 minutes, in which you discuss one use case and its data flow.
Frequently asked questions
Can companies use AI services from US providers under the GDPR?
Companies may use AI services from US providers when a lawful transfer mechanism exists for personal data and the contract with the provider meets GDPR requirements. Lawful mechanisms are certification of the provider under the EU-US Data Privacy Framework or standard contractual clauses with a transfer impact assessment. The legal department assesses the specific configuration.
What is the EU-US Data Privacy Framework?
The Data Privacy Framework is the arrangement to which the European Commission's 2023 adequacy decision refers. US companies can certify under it and may then receive personal data from the EU without further safeguards. Certification applies per company and can be checked in a public list.
Is the Data Privacy Framework still valid?
The Data Privacy Framework remains valid until the Court of Justice annuls the adequacy decision or the Commission amends it. The Commission reviews it periodically. Companies use it and additionally agree standard contractual clauses as a fallback.
What happens if the Data Privacy Framework is invalidated (Schrems III)?
Transfers that rely solely on the framework lose their legal basis with the judgment; its predecessor's invalidation came without a transition period. Companies with standard contractual clauses and a documented transfer impact assessment continue on that basis. The clauses therefore belong in every contract with a US provider today.
Are OpenAI, Microsoft, Google and Anthropic certified under the Data Privacy Framework?
Each provider's status is shown in the US Department of Commerce's public list at dataprivacyframework.gov. Check the legal entity with which you sign the contract and record the date of the check. Where certification is missing, base the transfer on standard contractual clauses with a transfer impact assessment.
What are standard contractual clauses?
Standard contractual clauses are contract terms drafted by the European Commission that bind data exporter and data importer to protective obligations. They serve as a transfer mechanism where no adequacy decision applies. A documented transfer impact assessment that evaluates the law of the destination country is also mandatory.
What is a transfer impact assessment?
A transfer impact assessment is the documented evaluation of whether the law of the destination country can override the safeguards of the standard contractual clauses. It looks at access by authorities and legal redress in the destination country as well as the supplementary measures. The standard contractual clauses require it.
Does the transfer review also apply to non-personal data?
The rules of Chapter V of the GDPR concern personal data only. Trade secrets and confidential documents fall under the EU Data Act, with protective obligations for cloud providers, as well as confidentiality agreements. One review framework covers both kinds of data in a single procedure.
Let’s talk about your initiative.
In 30 minutes we clarify where your organization stands and which step will pay off most now.
Request a consultation30 minutes · free · no obligation



