AI and the GDPR in accountancy: what to look out for

Can an accounting firm use AI under the GDPR?

Yes, the GDPR (in Dutch, the AVG) doesn't prohibit AI, but it sets requirements for how personal data is processed. Invoices, receipts and bank transactions often contain personal data, such as names, addresses and IBANs. As a firm you determine what that data is processed for, and an AI supplier processes it on your behalf. That calls for clear agreements and appropriate security.

The practical questions are therefore: what does the supplier use the data for, is it used to train AI models, who has access, how is the data secured and what happens when you stop. As a firm you also want to be able to explain how a booking came about, especially when the AI books on its own.

What should you agree with an AI supplier under the GDPR?

With an AI supplier you agree at least a data processing agreement, or DPA, stating which data is processed, for what purpose and under which security measures. Also ask about the subprocessors the supplier uses, its policy on training AI models and what happens to your data on termination. Those answers belong in writing.

Autoboeker offers a DPA agreement. Policies, subprocessors and security documentation are in the Trust Center at trust.autoboeker.nl. For procurement questions, the team is happy to walk you through the measures, the DPA and the subprocessors. That way you can check the agreements against your own firm's privacy policy before any client data is processed.

1. 1Sign a data processing agreement (DPA) with the supplier
2. 2Request the list of subprocessors
3. 3Check whether your data is used to train AI models
4. 4Check encryption, access management and logging
5. 5Record what happens to the data when you end the service

How does Autoboeker handle your clients' data?

Autoboeker uses your firm's and your clients' data only to deliver the service. The data is never sold or shared, not used to train public AI models and deleted when the account ends. The AI learning from corrections and history happens per administration and firm-wide within your own environment. Your data stays yours.

Technically, data is encrypted in transit with TLS and at rest with AES-256, with secure key management. Access is role-based and can be set per administration, and user activity is logged. There are also daily backups, redundant infrastructure, monitoring, incident response procedures and restricted access to production. Continuity and disaster recovery procedures are in place too.

• Never sold or shared
• Used only to deliver the service
• Not used to train public AI models
• Deleted when the account ends
• Encrypted with TLS in transit and AES-256 at rest
• Role-based and administration-specific access

How do you stay in control of what the AI books?

You stay in control with a confidence threshold per firm or administration: below it, a booking stays a proposal that a person reviews. You also specify which amounts, GL accounts or actions may never be automated. With approval before or after booking you decide who approves what, and with roles and permissions who sees which administrations.

Every booking shows a confidence score, the sources and the rule that was applied, so the AI is not a black box. Every action by users and AI is recorded in the audit log and can always be adjusted by hand. That way you can show afterwards how a booking came about and who approved it.

What about agentic AI such as Atlas?

Agentic AI such as Atlas handles tasks on its own, but only within the boundaries your firm sets. Atlas reviews bookings, processes invoices, resolves client questions and matches bank transactions. Your team sets the thresholds, the rules and which actions always require a human, and Atlas escalates whatever needs human judgement.

With Atlas, too, every automatic action is traceable and explainable, with the same audit log as the rest of the platform. Atlas doesn't replace the accounting software; it works on top of it and writes the result back. Atlas is available for firms from 25 FTE, in the Large firm plan with an SLA and DPA.

Is Autoboeker certified for information security?

Autoboeker is developed using security principles from ISO/IEC 27001, SOC 2 and the GDPR; that is not the same as a certification. The current certification status and documentation are in the Trust Center. If ISO certification of the supplier is a hard requirement for your firm, check this in advance so you don't have the wrong expectations.

Alongside the technical measures, Autoboeker works with periodic security reviews, secure coding standards, input validation, rate limiting and security awareness within the team. Single Sign-On and a 99.9% uptime SLA are available on request. You can go through security questions for your procurement in a call with the team, including the DPA and subprocessors.