AI is evolving to the point where it is capable of planning, deciding and executing actions autonomously. This shift introduces a new scenario with significant implications for the protection of personal data.
To assist organisations, the AEPD has published the guide “Agentic artificial intelligence from a data protection perspective”, identifying risks and offering recommendations aligned with the principles set out in Article 5 of the GDPR.
But… what do we know about agentic AI?
Agentic AI is an artificial intelligence system designed to act autonomously, combining analysis, planning and execution. These agents can reason, break objectives down into sub‑tasks, decide which steps to take and use different tools or sources of information.
This means that their ability to operate with a certain degree of independence increases the automation of data processing and makes it necessary to review how associated risks are managed.
So… how can agentic AI affect our company?
One of the main risks is the lack of predictability: by deciding for itself which actions to carry out, the AI may perform processing activities that were not initially envisaged, which can affect the principle of purpose limitation (Art. 5.1.b GDPR).
There is also a risk of infringing the data minimisation principle (Art. 5.1.c GDPR), as the AI may access external services, internal databases or third‑party information, expanding the volume of data processed beyond what is necessary.
As a result, data security may be affected; in other words, these autonomous systems may be vulnerable to attacks or failures that compromise data confidentiality and integrity, requiring appropriate technical and organisational measures in accordance with Article 32 of the GDPR.
In addition, their technical complexity makes it difficult to clearly track which decisions are made, which data are used and why, complicating the application of the accountability principle, which requires controllers to be able to justify such use (Art. 5.2 GDPR).
What does the AEPD recommend when using agentic AI in our company?
The AEPD recommends applying the principle of data protection by design and by default, incorporating privacy‑enhancing measures from the system’s development stage to limit processing to what is strictly necessary. For example, enabling the exercise of data subject rights, the reversal of actions, or defining the level of autonomy granted to the system (Art. 25 GDPR).
It also advises clearly defining processing responsibilities, identifying controllers and processors, especially where the agent uses third‑party tools or services such as Copilot.
Moreover, it suggests that, if changes such as new purposes or recipients are introduced, data subjects must be informed in order to comply with the transparency principle.
In any case, it should be borne in mind that a data protection impact assessment may be required prior to implementation where the use of the agent is likely to significantly affect individuals’ rights and freedoms (Art. 35 GDPR). An example would be its use to classify medical records according to treatments and outcomes.
In summary, the Agency emphasises the need to limit the agent’s permissions, segment access rights and review privileges to prevent it from operating on processing activities unrelated to its function.
In addition to all of the above, the AEPD considers continuous human oversight to be the main preventive factor. Such oversight should be carried out by qualified professionals, such as data protection officers or advisors, to avoid misalignment, bias and feedback effects.
In this field, PONTI & PARTNERS offers comprehensive support to ensure robust supervision and reduce unnecessary risks by implementing data protection policies tailored to each organisation. To this end, we bring solid experience in regulatory compliance, audits, risk assessment and the design of AI governance policies, enabling us to guarantee a safe, responsible implementation fully aligned with current regulations.
Article by Alba Gómez.
