Artificial Intelligence and Privacy: How to Protect Personal Data

Artificial intelligence can analyze large amounts of information, identify correlations, make predictions and generate new content. These capabilities are opening up important opportunities for individuals and businesses, but they also raise a fundamental question: what happens when personal data is among the information processed by an AI system?

The relationship between artificial intelligence and privacy arises precisely from this question. To function, many AI systems use large amounts of data during development, training or use. This data may include information relating to identifiable people, such as names, images, voice recordings, location, preferences, online behavior or professional information.

The presence of personal data does not in itself make the use of artificial intelligence problematic. However, processing must comply with specific principles, starting with those established by the General Data Protection Regulation (GDPR):

• lawfulness,
• fairness and transparency,
• purpose limitation,
• data minimization,
• accuracy,
• storage limitation,
• integrity and confidentiality.

These obligations are now complemented by the European Artificial Intelligence Regulation, or AI Act, which entered into force on 1 August 2024. The regulation introduces a common framework for the development and use of AI in the European Union, based on the risk level of different applications, and sets specific obligations for developers and users of certain systems.

The two regulations do not replace one another: when an AI system processes personal data, the provisions of the AI Act must be considered together with those of the GDPR. It is through this integration of innovation, data governance and protection of rights that truly responsible use of artificial intelligence can be built.

Artificial intelligence and privacy: what are the risks to personal data?

Privacy risks do not arise from artificial intelligence as such, but from the way data is collected, used, stored and combined within systems.

One of the first issues concerns the amount of information used. AI models can be developed using very large datasets from different sources. These collections may include personal data provided directly by users, obtained from public sources or supplied by third parties.

This makes it important to establish from the outset which data is actually necessary for the purpose of the system. The GDPR‘s data-minimization principle requires personal data to be adequate, relevant and limited to what is necessary for the purposes for which it is processed.

A second risk concerns the possibility of using information for purposes other than those originally intended. Data collected to provide a specific service, for example, cannot automatically be reused to train an AI model without verifying that the new processing is compatible with the original purpose and has an appropriate legal basis.

The issue becomes particularly complex with large-scale models. In December 2024, the European Data Protection Board (EDPB) published a specific opinion on the use of personal data in the development and deployment of AI models, addressing, among other things, the conditions for considering them anonymous, the use of legitimate interest as a possible legal basis, and the consequences of using unlawfully processed data during model development.

Anonymization deserves particular attention. According to the EDPB, it is not enough simply to state that a model is anonymous: the assessment must be made on a case-by-case basis, and it must be very unlikely both that the people whose data was used can be identified directly or indirectly and that such information can be extracted from the model through queries.

Alongside data collection, there are also risks linked to the use of outputs generated by artificial intelligence.

Profiling, bias and misuse of data

Profiling consists of the automated use of personal data to evaluate certain aspects relating to a person, such as preferences, interests, behavior, work performance, economic situation or reliability.

AI systems can make these analyses much more sophisticated by combining large amounts of information and identifying correlations that are difficult to detect manually. This can be useful for personalizing a service or identifying anomalies, but it can also increase the risk that people are classified on the basis of characteristics they are unaware of or that do not accurately reflect their situation.

A closely related issue is algorithmic bias. An AI system learns from the data used to build and train it. If this data reflects imbalances, errors or unrepresentative views of reality, the model may reproduce or amplify those distortions.

The issue is particularly important when AI outputs are used to support decisions about people. The GDPR provides specific safeguards for decisions based solely on automated processing, including profiling, when they produce legal effects or similarly significantly affect the individual. Where required by law, appropriate safeguards must be provided, namely:

• the possibility of obtaining human intervention,
• the possibility of expressing one’s point of view,
• the possibility of contesting the decision.

Another risk concerns the misuse of personal data. Information entered unknowingly by users into AI tools may include confidential documents, company information, personal data of customers or colleagues, and other content that should not be transferred to external services without prior assessment.

Generative AI also introduces a new dimension. Images, voices and videos available online can be used to create highly realistic artificial content. Deepfakes, for example, use deep-learning techniques to reproduce a person’s face and voice and can be used for impersonation, fraud and information manipulation.

AI and privacy therefore concern not only the confidentiality of information, but also the way data can be interpreted, combined and transformed through algorithms.

GDPR and artificial intelligence: what obligations apply to companies?

For a company using AI systems, the first principle to keep in mind is that the GDPR continues to apply fully whenever personal data is processed.

The introduction of artificial intelligence therefore does not create an exemption from data-protection rules. On the contrary, the complexity of these systems can make it even more important to understand precisely what information is being used, for what purpose, through which providers and for how long.

A first step is to identify a valid legal basis for the processing. Depending on the context, this may be, for example, performance of a contract, compliance with a legal obligation, consent or legitimate interest, where the conditions laid down by the regulation are met.

With regard to legitimate interest in particular, the EDPB has emphasized the need for a structured assessment. For AI models, organizations must verify the existence of a legitimate interest, the necessity of the processing in relation to that interest, and the balance with the rights and freedoms of the people involved.

Alongside lawfulness, companies must respect the principle of transparency. People must be able to understand what data is collected and for what purposes, in accordance with the requirements of the GDPR.

This requirement can become complex in more advanced AI systems, where the internal workings of models are not always immediately interpretable. Technical complexity, however, does not remove the obligation to provide data subjects with clear and understandable information about the processing.

Data processing and the accountability principle

One of the central principles of the GDPR is accountability. The data controller must not only comply with the law, but must also be able to demonstrate compliance through appropriate measures, procedures and documentation.

In the case of artificial intelligence, this means adopting governance that covers the entire lifecycle of the system: from selecting the provider and the data to be used through to operation and monitoring of outputs.

The GDPR also introduces the principles of privacy by design and by default. Data protection should therefore not be added after a service has been developed, but integrated from the design stage. Article 25 requires appropriate technical and organizational measures, such as pseudonymization and minimization, and establishes that, by default, only personal data necessary for the specific purpose should be processed.

Where processing may result in a high risk to people’s rights and freedoms, a Data Protection Impact Assessment (DPIA) may also be required. In AI projects, this analysis can be particularly important where large amounts of personal data, special categories of information, monitoring systems or tools supporting significant decisions about individuals are involved.

The relationship between AI and the GDPR can be summarized through a number of key operational principles:

The AI Act adds a further layer of governance. The European regulation follows a risk-based approach and sets different requirements depending on the type and use of the AI system. Compliance with the AI Act does not replace privacy obligations: where personal data is processed, the two sets of rules must be applied in a coordinated way.

AI and privacy: best practices for responsible use of data

Responsible use of AI starts with an apparently simple question: which data is really necessary? Reducing the amount of personal information used lowers exposure to risk and makes it easier to comply with the GDPR‘s data-minimization principle.

The same logic should be applied when choosing tools. Before using an AI platform, especially in a business environment, it is important to understand where data is processed, how it is stored, whether it can be used to further train the service, and which parties may have access to it.

For employees, this means avoiding the use of unauthorized AI tools for:

• confidential information,
• internal documents,
• personal data relating to customers and colleagues,
• credentials,
• content protected by confidentiality obligations.

Companies can support these rules with technological and organizational measures. Access-management systems, information classification, encryption, pseudonymization, activity logging and control procedures can help reduce the risk of misuse.

Alongside technical protection, it is essential to verify data quality. Inaccurate or incomplete information can produce equally problematic outputs. This principle is particularly important when AI is used in business processes that support decisions concerning people, customers or employees.

Human oversight also remains essential. An output produced by an algorithm should not be considered correct simply because it comes from an AI system. Information may be incomplete, inaccurate or influenced by the data used during model development.

For companies, a responsible approach can therefore combine several measures:

• preliminary risk assessment,
• careful selection of providers,
• data minimization,
• access protection,
• verification of outputs,
• employee training and continuous monitoring.

These elements become even more important in organizations that manage critical infrastructure, where digital innovation, information availability and security must progress together.

For Italgas, protecting information assets is an integral part of the digital-transformation journey. The Group oversees cybersecurity through a Next Generation SOC operating 24 hours a day and Cyber Threat Intelligence services designed to identify in advance threats targeting digital resources, information and corporate reputation.

Security becomes even more important as AI is increasingly integrated into business processes. Algorithms and models can generate value only if they use data that is reliable, protected and properly governed. Cybersecurity and privacy therefore become two complementary components of the same strategy: the former protects systems and information from threats, while the latter ensures that personal data is processed in respect of people’s rights.

Phenomena such as deepfakes also show how close the relationship has become between artificial intelligence, digital identity and information protection. Technologies capable of generating highly realistic images and voices make it increasingly important to verify the origin of content and protect the data that can be used to reproduce a person’s identity.

The challenge, therefore, is not to limit the use of artificial intelligence, but to create the conditions for adopting it consciously. Privacy by design, cybersecurity, data quality, transparency and human oversight make it possible to turn data protection from a constraint into a structural element of innovation.

From this perspective, the relationship between AI and privacy becomes a matter of trust. The more artificial intelligence enters business processes and everyday services, the more important it will be to understand how information is processed and to know that appropriate rules and systems exist to protect it. The sustainability of digital transformation will be measured by the ability to combine innovation and responsibility.