👉 GET THE AI USE POLICY SAMPLE
This analysis is intended primarily for employers and managers whose businesses use or plan to use artificial intelligence in everyday work.
Integrating generative AI and large language models into daily processes has become an economic reality. AI can significantly improve efficiency, but legal rules and internal policies have lagged behind technological development.
By 2026, the focus has shifted from efficiency alone to compliance. Alongside intellectual property and data protection risks, Estonian businesses must consider the EU AI Act, which places obligations on employers.
This analysis identifies four critical areas—confidentiality, intellectual property, data protection and regulatory compliance—and proposes a legal framework to manage them.
1. Irreversible disclosure of trade secrets
A major, often underestimated risk is losing control of company information. Many public generative AI platforms, including standard configurations of ChatGPT, Claude and Gemini, reserve rights in their terms to use submitted information for model training.
This can conflict with trade secret protection. A defining requirement is that the holder has taken reasonable measures to keep information secret. An employee entering a confidential contract, strategy or financial data into a public chat may be viewed as voluntarily disclosing the secret or breaching the duty of care.
Risk: In litigation, the company may lose the ability to rely on trade secret infringement because it failed, through its employee, to preserve confidentiality.
2. Compliance with the EU AI Act
The AI Act regulates both developers and businesses using AI systems as deployers. Two principles are particularly important in employment:
- Human oversight: AI systems should not make decisions without human control. Employers must ensure that employees critically assess generated output and take responsibility for the result. “The AI made a mistake” is not a legally sufficient explanation.
- AI literacy: Employers must ensure that staff using AI at work have adequate competence to understand its operation and limitations.
Without clear internal rules, these requirements may be breached, potentially triggering supervisory proceedings. This calls for a clear allocation of responsibility, rather than an outright ban on AI.
3. Uncertainty over intellectual property ownership
Copyright law and case law start from the principle that only a natural person can be an author. AI-generated output does not automatically qualify as a copyright-protected work.
If no protected work arises, copyright cannot be assigned. Why are contractual rules still necessary?
Without clear rules, a gap remains. A suitable AI policy addresses it on two levels:
- Hybrid protection: Most outputs combine human and machine contributions. The policy ensures rights arising from human creativity pass to the employer.
- Contractual assets: Even if generated code or text lacks copyright protection, the policy identifies it as an employer asset, such as a trade secret or know-how. It prevents employees taking it to a competitor through employment obligations rather than copyright claims.
Without this clause, an employee may argue that AI-generated material is public property they can freely reuse.
4. Irreversible GDPR breaches
Businesses sometimes assume that omitting a customer's name makes AI use safe. The position is more complex. Using public AI models, such as a free ChatGPT account, to process personal data can create three serious risks:
- Transfer without a legal basis. Entering customer data, including indirect identifying information, transmits it to a provider such as OpenAI or Google. The source analysis treats that provider as a third party under the GDPR and notes that free accounts may lack a company data processing agreement (DPA), leaving inadequate legal safeguards and control over use of the data.
- Difficulty fulfilling the right to erasure under Article 17. If personal data are used to train a model, they may become part of its learned parameters and cannot necessarily be selectively removed. The company may create a situation where it cannot technically fulfil an erasure obligation.
- Security and server location. Some providers process data outside the European Economic Area, including in the United States. Without appropriate safeguards, this can constitute an unlawful third-country transfer.
5. Conclusions and recommendations
A complete workplace ban on AI is not sustainable if it prevents innovation and competitiveness. However, unregulated use is no longer an acceptable approach.
An employer allowing AI without a clear legal framework risks:
- A structural GDPR problem that may be technically difficult to remedy through deletion.
- Loss of trade secret protection through failure to exercise due care.
- Conflict with the directly applicable EU AI Act.
A specific AI use policy is needed to manage these risks.
Does your company have AI rules in place?
We have prepared a sample designed around Estonian law and the AI Act.
👉 GET THE AI USE POLICY SAMPLE

