Skip to main content
Guides

Data sovereignty in AI sales: who owns your conversations?

Who can access your sales conversations, and how do AI tools use the data? Review contracts, data paths, export and the relevant EU rules.

Author: Agent360 editorial team
Published: 4 February 2026
Last updated: 23 October 2026
Reading time: 19 min read

Who owns your sales conversations?

Data sovereignty is used here to mean control over how your sales data is stored, processed and accessed. For recordings, transcripts and AI analyses, it is relevant to look at both the contracts and the actual data path.

Sales conversations can contain prices, customer needs and confidential business information. Before you send them to an AI tool, you should know who receives them, what they are used for, and how they can be exported or deleted.

This article goes through questions for your AI infrastructure. The vendor's nationality is not, on its own, an answer to whether a solution meets your requirements.

Which rights and access arrangements should you clarify?

TL;DR:

  • Look into rights, access, storage and export for both conversations and AI output.
  • Assess possible access by public authorities on a case-by-case basis; an EU region alone is not sufficient documentation.
  • Assess hosting, access and the supply chain together before you choose an AI solution.

Timeline updated: October 2026

Data sovereignty in practice Distinguish between rights to data, practical control over it, and the rules the processing is subject to. Server location, support access, sub-processors and contracts must be assessed together.

Three questions can structure the review:

  • Rights: What does the contract say about recordings, transcripts, analyses and any model training?
  • Operations: How can you export and delete data, and what limitations apply?
  • Access: Who can access which data, from where and on what terms?

Agent360 can build solutions around specific sales workflows. The requirements for data and access must be agreed and tested for each individual solution. See also our guide to AI in the sales department.

How do you examine the data path in an AI sales tool?

An AI tool can be hard to assess if all you see is the input and the result. So ask for documentation on three things:

  1. The data path: Which services receive audio, text and metadata? Does that also include support, logs and backups?
  2. The purpose: Is data used solely to deliver the feature, or also for model training and other purposes? Read the specific contract and the settings.
  3. The access: Which legal entities and people can access data? What procedures apply to requests from public authorities?

Test the feature you are buying, and check the documentation. A promise of efficiency doesn't answer those questions.

What does the US CLOUD Act mean for data ownership over European sales data?

The CLOUD Act clarifies in US law that the obligation of covered service providers to preserve or disclose information under the Stored Communications Act can apply to data in their possession, custody or control, even when the data is stored outside the US. It is not a general right of access for every US authority to every company's data. See the US statutory text, § 2713.

For your assessment, that means the server's country can't stand alone. Look into the specific supply chain, access arrangements and possible transfers. The Danish Data Protection Agency's cloud FAQ states that a US provider can be used without a transfer to the US when that has been agreed and the controller is satisfied that the provider complies with applicable EU law. The risk to data subjects' rights in the event of a possible disclosure must still be assessed.

Question Documentation to ask for
Where is data processed? Data flow showing services, regions, support and sub-processors
Who has access? Access roles and a description of key management
Is data used for model training? The applicable contract, settings and delimitation of purpose
How are requests from public authorities handled? The vendor's procedure and relevant contract terms
Can we change vendor? Export format, scope, deadlines and any fees

Use the same questions for Agent360 and other vendors. A general comparison can't replace the documentation for the specific service and configuration.

What is Schrems II, and is it still relevant in 2026?

The Schrems II ruling from 2020 declared the EU-US Privacy Shield invalid. The new EU-US Data Privacy Framework (DPF) is based on the Commission's adequacy decision of 10 July 2023, not on a court ruling from 2023. The DPF can be used as a transfer basis for US companies that participate in the scheme. Businesses must still comply with the other GDPR requirements; see the Danish Data Protection Agency's rules and guidance and the Commission's DPF overview.

The General Court dismissed Philippe Latombe's action for annulment on 3 September 2025 in case T-553/23. Latombe appealed the judgment on 31 October 2025 in case C-703/25 P. These procedural details establish neither a particular outcome nor a date for any change to the DPF. See the General Court's press release and InfoCuria's case overview.

For Danish sales departments, that creates three concrete challenges:

  • Transfer basis: Check that the specific transfer is covered by a valid basis, for example the DPF for participating companies or standard contractual clauses with the necessary conditions and measures
  • Ongoing assessment: You need to follow the legal framework and the specific data transfers. An EU-based vendor does not in itself remove all risks; access and the whole supply chain must also be examined.
  • Documentation: Keep the assessment and the contracts it is based on, so that changes in services or access can be reviewed.

For a solution you want built, the requirements can be framed as questions before you choose a vendor:

Area What needs to be clarified?
Legal basis for processing Which basis fits the specific processing?
Data volume Which information is necessary for the task?
Hosting and access Where is data processed, and who can access it?
Transparency What information must customers and employees be given?
Deletion Which data is deleted when, and how are backups handled?

What does the EU AI Act require of sales tools with voice analysis?

The EU AI Act's duty to inform on AI interaction has applied since 2 August 2026, and the requirements for high-risk systems in Annex III apply from 2 December 2027.

Legality and risk class follow the specific function and use, not the label "sentiment analysis" alone. Emotion inference from biometric data in workplaces and educational institutions is covered by the prohibition in Article 5(1)(f), with exceptions for medical or safety purposes. For lawful systems that, after a specific assessment, are high-risk, the following requirements may be relevant depending on the company's role:

  • Risk management: Systematic identification and handling of risks in the AI system
  • Data governance: Quality requirements and governance for training, validation and test datasets, where they are used. For systems without model training, the relevant requirements apply to the test datasets alone
  • Human oversight: The system must be effectively overseen by people. Oversight measures must be appropriate to the risk, autonomy and use, and make it possible, in an appropriate and proportionate way, to decide not to use the system, to disregard or reverse output, and to intervene or stop the system safely
  • Fundamental Rights Impact Assessment (FRIA): Mandatory only for the deployers and uses covered by Article 27: bodies governed by public law and private entities providing public services, for high-risk AI under Article 6(2), as well as deployers of systems for creditworthiness and for risk assessment and pricing in life and health insurance under Annex III, points 5(b) and 5(c). Systems in Annex III, point 2, are exempt. The requirement does not follow automatically from a tool being used for sales.
  • Technical documentation: A detailed description of the system's design and performance

The fine framework is up to EUR 35 million or 7% of global annual turnover for violations of the prohibitions in Article 5. For the obligations listed in Article 99(4), including Article 50 on transparency, the framework is up to EUR 15 million or 3%. The percentage caps are calculated on global annual turnover in the preceding financial year. As a starting point the higher cap is used; SMEs have the lower one in both categories, and small mid-caps have the lower one under paragraph 4. The specific fine depends on the circumstances of the case. See Article 99. The postponement has been adopted: under Regulation (EU) 2026/1744 of 8 July 2026, the requirements for high-risk systems in Annex III apply from 2 December 2027. The duty to inform on AI interaction, on the other hand, has applied since 2 August 2026.

For a new solution, you should clarify the relevant rules and requirements before choosing the data path and the vendors. A product name or a general compliance statement is no substitute for assessing the specific use.

Which risks should you assess when processing sales data?

A lack of overview can make it hard to document the data processing and to move a solution later. Assess the following, among other things:

Data protection: Check the legal basis, access and any transfers to third countries. A transfer can be lawful when the relevant conditions are met.

EU AI Act risk: The duty to inform on AI interaction has applied since 2 August 2026, and the high-risk requirements in Annex III apply from 2 December 2027. Violations of Article 50 on transparency carry a fine framework of up to EUR 15 million or 3% of global annual turnover. The framework of EUR 35 million or 7% applies to the prohibitions in Article 5. The company's category and the circumstances of the case must be taken into account; turnover alone can't be used to calculate a specific fine.

Changing vendor: Clarify export and migration before the contract is signed. The Data Act entered into force on 11 January 2024 and has applied since 12 September 2025. Chapter VI requires covered providers of data processing services to remove obstacles to switching and to support export of the data covered. It is not an unconditional right to all data or to the vendor's protected intellectual property rights and trade secrets. Until 12 January 2027, cost-based switching charges may still apply. See the Commission's explanation of the Data Act.

Confidentiality: Clarify whether data may be used for purposes other than the agreed function, including model training. Don't infer training rights from a vague term such as "product improvement" alone; read the applicable terms.

What needs to be clarified in a solution built with Agent360?

Agent360 can build solutions for specific sales workflows. The following are design questions for such a solution, not a list of standard features or operating guarantees.

Storing audio and analyses

Agree what is to be stored and for how long. For sales training and coaching, for example, audio, transcription, analysis and logs must all be part of the clarification. A requirement for no lasting audio storage must be tested across the whole data path.

Keys and access

Clarify who holds the encryption keys and can decrypt data. A customer's own API key for an AI service is not the same as customer-controlled encryption of stored data. A promise that the operator can't read the data requires separate documentation.

Transcription and language

Choose the service and operating model based on the specific requirements. Test the quality on relevant, lawfully usable examples with Danish language and dialects, and examine the whole supply chain. A European product name doesn't document where processing takes place or what the quality is.

Any model training

Clarify whether training is needed at all, which data it would use if so, and which rights and access limits are agreed. A separate customer account does not in itself document a separately trained model.

How do you assess the data flow in your own industry?

Start with the specific information and uses. An industry label alone doesn't determine which requirements a solution must meet.

Use Questions for the clarification
Advisory conversations Which confidential or personal information can be involved?
Recruitment and employee support How are the analyses used in decisions about people?
Customer service and sales What is recorded, and what are customers told?
Product and quote work Which business information may be shared with external services?

Describe the actual data and decisions before you choose the technology. Have the relevant requirements assessed for the specific solution.

What should a vendor be able to document with a customer case?

A relevant customer case should make it possible to tell a demonstration from a solution in operation. Ask for a precise description of the delivery, the data path and the measured result, and for permission to use the customer as a reference.

If the case includes CRM updating, it should, for example, show which information is transferred and how errors are handled. An anonymous quote alone documents neither the functionality nor the data protection.

Seven questions to ask your current AI sales vendor

Ask for answers and documents that apply to the specific service and contract:

  1. Where is data processed, including support, logs and backups?
  2. Which legal entities can gain access, and how are requests from public authorities handled?
  3. May data be used for model training or other purposes, and which settings apply?
  4. What is deleted on termination, when, and with what exceptions?
  5. Which data can be exported, in what format and at what cost?
  6. What does the contract say about rights to data, output and any customer-specific models?
  7. How is responsibility for informing people about AI interaction divided?

What can control over data be used for in sales work?

Access to your own recordings and analyses can support follow-up, reporting and changes to workflows. Agree export and integration based on the actual need, and test whether the data can be used in the system you want to move it to.

Agent360 can build a specific integration once scope, access and vendor terms have been clarified. This is not a general promise of access to all data or of rights to any model. Also read our guide to AI infrastructure and SaaS.

Which EU rules should you keep separate in the assessment?

Keep these sets of rules, among others, separate in the assessment:

The EU AI Act (duty to inform on AI interaction applicable since 2 August 2026): The requirements for high-risk AI systems in Annex III become binding on 2 December 2027, postponed by Regulation (EU) 2026/1744. Voice analysis must be assessed according to function and use, including whether the rules on emotion inference are relevant. Article 50 on transparency falls under the fine framework of EUR 15 million or 3% of global annual turnover; EUR 35 million or 7% applies to the prohibitions in Article 5. Special rules on company size and the specific case must be taken into account.

The EU Data Act (applicable since 12 September 2025): The regulation entered into force on 11 January 2024. Chapter VI sets requirements for switching between covered data processing services and for export of the data covered. The rules have limits and do not give an unconditional right to all data or to the vendor's protected intellectual property rights and trade secrets.

Conclusion: describe the actual data path

Gather contracts, access arrangements and technical documentation before you make promises to customers about their data. Use specific, delimited descriptions that match the solution you actually use.

Contact Agent360 if you'd like to discuss the data flow in a specific AI workflow.

Q: What is data sovereignty in practice for a sales department?

Data sovereignty here is about control over sales data: who can gain access, what processing takes place, and how data can be deleted or moved. EU hosting or a European legal entity is not in itself a guarantee of this control. Assess the specific processing, access from other countries, the whole supply chain, the contracts and the relevant rules on any transfers to third countries.

Q: Are American AI sales tools illegal in the EU?

An American vendor is not in itself incompatible with GDPR. Assess the contracts, the processing, the access arrangements and the whole supply chain. Transfers to the US require a relevant basis under GDPR's rules on third-country transfers. The DPF can be used for covered participants; other mechanisms, including standard contractual clauses, have their own conditions. A transfer basis is not a general approval of an AI product.

Q: What is the difference between the CLOUD Act and GDPR?

GDPR regulates the processing of personal data. The CLOUD Act clarifies in US law that the obligation of covered service providers to preserve or disclose information under the Stored Communications Act applies even when data in their possession, custody or control is stored outside the US. GDPR requires a valid legal basis for processing. Consent is one of the bases in Article 6; contract, legal obligation, vital interests, a task carried out in the public interest or in the exercise of official authority, and legitimate interests are other options under their conditions. American legislation can give rise to risks in connection with access by public authorities, but an American vendor is not in itself incompatible with GDPR. The contracts, the specific processing, the whole supply chain and any transfer basis must be assessed. See the Commission's overview of legal bases for processing and the Danish Data Protection Agency's cloud guidance.

Q: Can I just choose an "EU region" with my American cloud vendor?

An EU region does not on its own document that all processing and access take place in the EU. Also look into support access, sub-processors and any transfers to third countries. The Danish Data Protection Agency's cloud FAQ states that a US provider can be used without a transfer to the US when that has been agreed and the controller is satisfied that the provider complies with applicable EU law. The risk to data subjects' rights in the event of a possible disclosure must still be assessed. The vendor's home country therefore does not decide the question on its own, and a European vendor is not an automatic guarantee either. See the Danish Data Protection Agency's cloud guidance.

Q: What does it cost to switch from an American to a European AI platform?

It depends on data volume, export options, integrations and the requirements for the transition. Ask for a specific scope and an estimate. Chapter VI of the Data Act has applied since 12 September 2025 and sets requirements for switching and export, with limits laid down by law; it does not set a single price or a general delivery time for your migration.

Q: What does the EU AI Act require of sales tools in 2026?

The EU AI Act's duty to inform on AI interaction has applied since 2 August 2026, and the requirements for high-risk systems in Annex III apply from 2 December 2027, postponed by Regulation (EU) 2026/1744 of 8 July 2026. For emotion inference from biometric data, you must first assess whether the use falls under the prohibition in workplaces and educational institutions, or is covered by an exception for medical or safety purposes. For lawful uses, the risk class must then be assessed specifically. Which requirements apply depends on the specific use and the company's role. A FRIA is mandatory only for the deployers and uses covered by Article 27, as described in the section above; the requirement does not apply automatically to all sales tools or all high-risk systems. Violations of Article 50 on transparency fall under the fine framework of EUR 15 million or 3% of global annual turnover. The framework of EUR 35 million or 7% applies to the prohibitions in Article 5, with special rules on company size and a specific assessment of the fine.

Q: What is Schrems II, and does it still apply in 2026?

Schrems II is the 2020 judgment of the Court of Justice of the EU that declared the EU-US Privacy Shield invalid because US surveillance laws conflict with GDPR. The Commission adopted the new DPF decision on 10 July 2023. The General Court dismissed Latombe's action for annulment on 3 September 2025, and the judgment was appealed on 31 October 2025 in case C-703/25 P. That gives no basis for fixing a date or a particular outcome for the further judicial review. The choice of vendor must be based on a specific assessment of processing, access, supply chain and any transfer basis; an EU address is not in itself sufficient documentation.

Q: How do we assess data protection in an Agent360 solution?

Go through the specific solution, the contracts, the data path and the division of responsibility. Ask for documentation of the features and operating conditions you need. A Danish vendor address, EU hosting or the label BYOK is not in itself documentation that the whole solution complies with GDPR.

Sources

Tags: data sovereignty, GDPR, compliance, AI strategy, cybersecurity, EU hosting, CLOUD Act, Schrems II

Unsure about the rules for your own set-up?

We look at what you use today and what it requires: documentation, consent and what the EU AI Act means for you. 20 minutes, no commitment.

How we built it · Sales & follow-up

Sales cockpit with AI

A call list for salespeople. The call is transcribed in Danish, AI suggests the outcome (e.g. “Call again”) and drafts the email based on what the customer said.

  1. The salesperson calls
  2. AI writes transcript and email
  3. The salesperson sends
See the call and the follow-up →