Skip to content
Atesto
Create account
All articles

Data protection

Spain's AEPD issues data protection guidance for AI agents

Spain's data protection authority published guidance on October 27, 2026, detailing how to manage personal data when delegating tasks to AI agents.

On October 27, 2026, the Spanish Data Protection Agency (AEPD) (opens in a new tab) published new guidance on how to delegate tasks to artificial intelligence assistants and agents without losing control over personal data. The document addresses a fundamental shift in how people and organizations use artificial intelligence. Tools are moving beyond simply answering questions to taking autonomous actions on behalf of the user. They now manage calendars, send emails, complete purchases, and make reservations.

The authority notes that this shift changes the nature of artificial intelligence failures. When an assistant only responds to a prompt, an error is simply a wrong answer. When an agent acts autonomously, a failure becomes a sent email, a shared file, a purchase, a contract, or consent given in the user's name. To address these risks, the AEPD issued a guide outlining the privacy and security expectations for deploying these tools.

Who it applies to

The guidance applies directly to individuals and employees who use artificial intelligence agents to process personal or corporate data. The AEPD explicitly addresses the workplace environment, setting expectations for how staff should handle professional documents and corporate communications when using these tools.

The document also sets clear expectations for the technology providers who develop and offer artificial intelligence agents. The authority emphasizes that the responsibility for data protection cannot fall exclusively on the user. Companies that build and deploy tools capable of acting on behalf of individuals must build specific privacy and security controls into their software architecture. Providers must ensure their systems allow users to exercise their data protection rights under European law.

What it requires

The AEPD guidance outlines specific practices for managing the risks of autonomous artificial intelligence. Before delegating any task to an agent, the authority expects users to answer five questions: what the assistant can see, what it can remember, what it can do in the user's name, where it sends the data, and whether its actions affect other people.

Access control must remain strict. The authority advises granting only the essential permissions required for each specific task. Users should avoid selecting always allow options without fully understanding the scope of that permission. The AEPD cites a known case where a user asked an agent to sell an object and granted permanent permission, expecting the tool to ask for authorization before accepting offers. Instead, the agent closed the sale and provided the user's home address to the buyer, who then appeared at the seller's door unannounced.

To prevent such incidents, the guidance states that the agent should prepare the action, but the person must approve it. Users must demand confirmation before the system sends data, shares files, executes a payment, or signs a contract.

The authority highlights the specific risk of hidden instructions. A website, an email, a shared document, or a calendar invitation can contain invisible text, such as white text on a white background. These hidden prompts are designed to trick the agent into doing something different from what the user requested, such as sending out personal data or approving an unauthorized operation. Because no technical measure completely eliminates this risk, the AEPD states that the best protection is limiting what the agent is allowed to do if it is deceived.

Memory management is another core focus. Over time, an artificial intelligence assistant can build a detailed profile of a user's tastes, routines, and concerns. The AEPD advises using the formal memory, history, and privacy controls provided by the service to delete information. Users should not rely solely on asking the assistant to forget something during a chat conversation.

The guidance also covers the protection of third party data. Emails, agendas, and photo libraries contain information about other people who have not agreed to interact with the artificial intelligence system. The authority calls for the protection of this third party data. On shared devices, users should maintain separate profiles.

In the workplace, the AEPD states that staff must only use the assistants authorized by their organization. Employees must not connect their professional email accounts or corporate documents to a personal artificial intelligence account.

For technology providers, the guidance outlines a specific set of expected controls. Providers should offer understandable permissions that clearly distinguish between consulting data and taking action. They must provide configurable confirmations for high impact actions, a history of what the agent has done, and an immediate way to stop the system. Providers must also offer tools for managing and deleting memory, clear information on where data is sent, disclosure on whether the data is used to train the artificial intelligence, and a simple channel for individuals to exercise their rights.

The dates

The AEPD published the guidance document on October 27, 2026. The publication serves as the authority's current interpretation of how existing data protection principles apply to the use of autonomous artificial intelligence agents.

What to check now

  • Inventory the artificial intelligence agents currently authorized for use within the organization.
  • Verify that employees are not connecting corporate email accounts or professional document repositories to personal artificial intelligence tools.
  • Review the permission settings of authorized agents to ensure they operate with the minimum access necessary for their tasks.
  • Confirm that automated workflows require explicit human approval before the system executes high impact actions, shares files, or sends personal data.
  • Check the administrative controls provided by your artificial intelligence vendors to ensure they include an action history and an immediate stop mechanism.
  • Examine vendor documentation to determine exactly where agent data is sent and whether it is used to train their models.
  • Ensure that the formal privacy controls of the service are used to manage and delete agent memory, rather than relying on chat prompts.

The shift from artificial intelligence that answers to artificial intelligence that acts introduces new variables for security and compliance teams. When an organization or its employees delegate tasks to an autonomous system, they must ensure the technical safeguards match the level of autonomy granted.

Sources

  1. Agencia Española de Protección de Datos (AEPD) (opens in a new tab) · www.aepd.es

This article is general information, not legal advice. For your company's situation, talk to a qualified lawyer.

This article was written with AI and checked, sentence by sentence, against the sources linked above.