Artificial intelligence and confidentiality: may a company enter client data into artificial intelligence tools?

Artificial intelligence and confidentiality: may a company enter client data into artificial intelligence tools?

1. Question and short answer

May a company enter client data into an artificial intelligence (AI) tool, in particular a generative AI tool based on large language models? Yes, but only if several conditions are met cumulatively. Yes, but only if several conditions are met cumulatively.

There is no general prohibition. However, entering data into a prompt is, as a rule, a processing operation and often a disclosure to a third party, and is therefore subject to Regulation (EU) 2016/679 (GDPR), Portuguese Law No. 58/2019, contractual and statutory confidentiality obligations and, where applicable, professional secrecy.

In practical terms, using free or consumer tools, without a specific contract and without guarantees on retention, reuse and location of data, to process identifiable client data will hardly ever be compliant. Professional solutions, contractually framed and combined with data minimisation and internal governance, can be lawful.

 

2. . Data protection framework

2.1 Processing and responsibility. Entering personal data into an AI tool is processing (Art. 4(2) GDPR). The company is, as a rule, the controller (Art. 4(7)) and must comply with, and be able to demonstrate compliance with, the principles of Art. 5: lawfulness, fairness and transparency, purpose limitation, data minimisation, storage limitation, integrity and confidentiality (Art. 5(2) and Art. 24). Data on legal persons is not, as such, personal data, but corporate client documents frequently contain data on individuals (shareholders, representatives, beneficial owners, employees).

2.2 Lawfulness and purpose. A lawful basis under Art. 6(1) is required (performance of a contract, legal obligation, legitimate interests, consent, among others), and the purpose must be compatible with the one for which the data was collected (Art. 5(1)(b) and Art. 6(4)). Legitimate interests (Art. 6(1)(f)) require a documented balancing test; consent is valid only if freely given, specific, informed and unambiguous (Arts. 4(11) and 7). Special categories of data (Art. 9, for example health data, racial or ethnic origin or religious beliefs) and data relating to criminal convictions (Art. 10) are subject to reinforced restrictions, so entering them into external tools should, as a rule, be avoided.

2.3 Transparency. Data subjects must be informed of the purposes, the recipients or categories of recipients and any transfers to third countries (Arts. 13 and 14). The privacy notice should therefore reflect the use of AI providers.

2.4 The tool provider. Where the provider processes data on behalf of and on the instructions of the company, it acts as a processor, and a contract or other binding legal act with the content required by Art. 28(3) is mandatory (documented instructions, confidentiality, security measures, sub-processing, assistance, deletion or return of data and audits). If the provider uses the data for its own purposes, such as training or improving its models, it may be considered a controller for that purpose (Art. 28(10)), which requires an independent lawful basis and information to data subjects. This is the main risk of general-purpose tools: enterprise versions and application programming interfaces (APIs) generally offer contractual guarantees that free or consumer versions do not, but this must be verified against each provider’s actual terms.

2.5 International transfers. If data is processed or accessible outside the European Economic Area, Arts. 44 to 49 apply: an adequacy decision (Art. 45), such as Implementing Decision (EU) 2023/1795 on the EU-US Data Privacy Framework, which applies only to certified entities, or appropriate safeguards such as standard contractual clauses (Art. 46), together with an assessment of the transfer risk in accordance with Schrems II (CJEU, Case C-311/18). The validity of the EU-US adequacy decision has been challenged in court, so its continued validity should be confirmed at the time of use.

2.6 Security, impact assessment and incidents. The company must adopt appropriate technical and organizational measures (Articles 25 and 32), maintain a record of processing activities (Article 30), and carry out a data protection impact assessment (Article 35) when processing is likely to involve a high risk, which may occur with the use of new technologies, on a large scale, or with special categories of data (cf. CNPD Regulation No. 1/2018, which establishes the list of processing activities subject to impact assessment). Unauthorised disclosure of data to a supplier, or unauthorized access to such data, may constitute a personal data breach (Article 4, paragraph 12), subject to notification to the CNPD, whenever possible within 72 hours, unless it is unlikely to result in a risk to data subjects (Article 33), and communication to data subjects when the risk is high (Article 34).

2.7 Penalties and regulatory guidance Infringements of the GDPR may result in fines of up to EUR 10 million or 2% of worldwide annual turnover (Art. 83(4)) or up to EUR 20 million or 4% (Art. 83(5)), depending on the provisions infringed, under the administrative offence regime of Law No. 58/2019, without prejudice to data subjects’ right to compensation (Art. 82). In Opinion 28/2024 of December 2024 on the processing of personal data in the context of AI models, the European Data Protection Board stressed the need for a case-by-case assessment, the role of legitimate interests and mitigating measures, and the demanding standard for treating a model as anonymous.

 

3. Confidentiality beyond the GDPR

3.1 Contractual duties. Irrespective of whether personal data is involved, client contracts frequently contain confidentiality clauses, and good faith in the performance of obligations (Art. 762(2) of the Portuguese Civil Code) may ground ancillary duties of secrecy. Disclosure to third parties, including an AI provider, without contractual authorisation may give rise to contractual liability (Art. 798 of the Civil Code), the obligor being liable for the acts of its auxiliaries (Art. 800). The specific clauses should be reviewed, including any restrictions on sub-contracting and on data location.

3.2 Trade secrets. Confidential business information about clients may constitute a trade secret, protected by the Industrial Property Code (Decree-Law No. 110/2018, articles 313 et seq.), which transposed Directive (EU) 2016/943. One of the requirements is that the information has been subject to reasonable due diligence to keep it secret; its introduction into tools without confidentiality guarantees may weaken this protection.

3.3 Professional secrecy and sector-specific confidentiality. If the company practices a profession or activity subject to secrecy, such as law (Article 92 of the Statute of the Bar Association, Law No. 145/2015) or banking (Article 78 of the RGICSF), the standard is more stringent. Disclosure of another's secret known by reason of the profession is punishable by law (Article 195 of the Penal Code) and may constitute a disciplinary offense. The introduction of information covered by secrecy into a third-party tool, without equivalent confidentiality guarantees, may be considered disclosure and should therefore be avoided, unless authorized by the client and with a solution that binds the provider to an appropriate duty of confidentiality. Specific guidelines issued by the respective professional body or sector regulator should also be consulted.

3.4 Workers. Those acting under the authority of the data controller may only process data on their instructions (Article 29 of the GDPR). In exercising their managerial power (Article 97 of the Labour Code), the employer must define clear rules on the use of AI, including prohibiting personal accounts for professional tasks.

 

4. Regulation of Artificial Intelligence

Regulation (EU) 2024/1689 (AI Regulation) does not override the GDPR or confidentiality obligations (Article 2(7)). The company using an AI system under its authority is "responsible for deployment" (Article 3(4)). Since 2 February 2025, the prohibited practices and the duty of AI literacy (Article 4), as amended by Regulation (EU) 2026/1744 (AI Digital Omnibus), published on 24 July 2026 and in force since 27 July 2026, apply, under which providers and those responsible for deployment must take measures to support the development of AI literacy among their staff.

The same decree deferred the obligations relating to high-risk systems in Annex III until 2 December 2027. If the tool is used in high-risk contexts (e.g., recruitment or evaluation of workers), then the requirements of Article 26 will apply, including human supervision and record keeping.

The proposed changes to the GDPR in the data component of the Digital Omnibus, specifically regarding the legitimate interest for AI development, had not yet been adopted at the reference date, and the GDPR remains applicable in its current wording.

5. Practical scenarios

Scenario	Assessment	Minimum requirements
Free or paid tool, personal account, identifiable customer data.	High risk; generally non-compliant	Do not enter customer data; use only fictitious or effectively anonymized data.
Business solution with a contract under Article 28, without training using company data and with location and transfer guarantees.	Allowable, subject to conditions.	Legal basis, information to data subjects, impact assessment where applicable, minimization, internal policy and access control.
Anonymized data	Outside the scope of the GDPR only if anonymization is effective.	Assess the means reasonably likely to allow for re-identification (recital 26); pseudonymization does not preclude the application of the GDPR (Article 4(5)).
Special categories of data, information under professional secrecy or strategic confidentiality.	Very high risk	As a rule, avoid this; if indispensable, obtain client authorization, a private solution with enhanced guarantees, and an impact assessment.

 

6. Recommendations
7. Map The tools used, the purposes for which they were used, and the categories of data entered, including any unauthorized uses.
8. Approve an internal AI policy.: authorized tools, prohibited data (special categories and classified information), prohibition of personal accounts, human verification of results, and disciplinary consequences.
9. Conduct due diligence with the supplier.Article 28 contract; contractual exclusion of training with company data; minimum retention; location and subcontractors; security measures and certifications; erasure and incident reporting.
10. Minimize and anonymize Before entering data, prefer using fictitious or synthetic data for testing and training.
11. Document The basis for legality, the balancing of legitimate interest, the impact assessment, the registration of activities, and the information provided to data subjects.
12. Review contracts with clients. (confidentiality, subcontracting) and, if necessary, obtain prior informed authorization.
13. Forming the teams (AI literacy, art. 4 of the AI Regulation) and implement technical controls: corporate accounts, data breach prevention and usage logs.
14. Prepare for incident response., with a procedure for assessing and reporting data breaches (Articles 33 and 34 of the GDPR).

 

7. Conclusion

The question is not whether AI can be used, but under what conditions. Companies that input customer data into AI tools must have a legal basis, an appropriate contract with the provider, guarantees regarding data transfers and reuse, respect for confidentiality obligations, and documented internal governance. In the absence of these conditions, data entry should be avoided or limited to fictitious or effectively anonymized data.

 

 

Notice. This notice is for informational and general purposes only, reflects the legal framework and publicly available information as of the reference date (September 18, 2026), and does not constitute legal advice, nor does it create a mandate or legal consultation relationship. The application of the rules depends on the specific circumstances of each case, the content of the contracts and the terms of use of the tools in question, and the regulatory framework is constantly evolving. Before making any decision, specialized legal advice should be obtained. and customized.